CN103886034B - A kind of method and apparatus of inquiry input information that establishing index and matching user - Google Patents

Abstract

The object of the present invention is to provide a kind of for establishing the method and apparatus of the inquiry input information of index and matching user；According to text information, determines structured message and extract descriptor；According to the corresponding theme of descriptor, corresponding label word is determined；It establishes and indexes for the descriptor and label word.Further, descriptor and label word are obtained to the inquiry input information analysis of user's input, and carries out matching inquiry in the index of aforementioned foundation accordingly, obtain candidate text information；According to the semantic matching degree of candidate text information and inquiry input information, the determining target text information to match with inquiry input information.Compared with prior art, the present invention is based on encyclopaedia class or other Internet resources knowledge, carry out the extraction of theme, title, form effective description to resources and knowledge content, so that more efficient to the semantic search of this kind of resources and knowledge, meet the complicated description search need that user can not accurately be reached using antistop list, improves the usage experience of user.

Description

A method and device for indexing and matching user's query input information

technical field

The present invention relates to the field of computer technology, and in particular, to a technology for establishing an index and matching query input information of a user.

Background technique

In the process of using search engines, people often do not know what keywords to enter to express their thoughts. They may enter a bunch of descriptive words, such as: 1) Vomiting in the morning, usually flustered and short of breath, weak in limbs, yes What disease symptoms? 2) Introduction to the song expressing nostalgia for the lover? 3) Contains the song "What to say to forget about wealth" 4) In which movie was eating hot pot and singing the song, and who said it? 4) Describe the verse of diligent study 5) It is difficult to be a man, and it is difficult to be a woman, who said it, and what is the complete statement? Some users may also enter some complex expressions. For example, for some character categories, users may ask "What are the emperors and state presidents from Anhui?", "Introduction to the Standing Committee of the Political Bureau of the current government in Shanxi" and so on. In this case, it is difficult for search engines to find suitable results.

From the analysis of the reason, this is because the current general search engine mainly builds an index on the title. Although these search engines usually index the content, due to factors such as weight adjustment, it is difficult to display some high-quality knowledge descriptions well. For example, for some resource categories such as songs, movies, etc., existing search engines usually only index the names of songs and movies. Existing search engines cannot make effective search queries when there are lyrics, introductions to lines, or small descriptions. These situations also occur in resources such as novels, poems, couplets, blessings, characters, TV series, novels, sentences, idioms, diseases, etc.

Encyclopedia resource knowledge is usually indexed with words as the center, which makes it difficult to rank keywords that do not appear in the title in the general search ranking algorithm. However, in fact, due to the knowledge authority of encyclopedia resource knowledge, if these data are ranked in the front, it can well meet the needs of users. For example, for diseases in the encyclopedia, if the symptoms are tagged and indexed, the corresponding resource knowledge can be well provided to the user according to the symptoms described by the user.

Therefore, how to effectively utilize the existing resource knowledge, establish an index for it, and obtain the target text information corresponding to the user's query input information by matching, has become one of the problems that those skilled in the art need to solve urgently.

SUMMARY OF THE INVENTION

An object of the present invention is to provide a method and apparatus for indexing and matching user's query input information.

According to one aspect of the present invention, there is provided a method for indexing based on textual information, wherein the method comprises the following steps:

A determines structured information from textual information;

B extracts subject headings from the structured information;

C, according to the theme corresponding to the theme word, from the text information, determine the label word corresponding to the theme;

D builds an index for the subject term and the tag term.

According to another aspect of the present invention, there is also provided a method for matching user's query input information according to the aforementioned established index, wherein the method comprises the following steps:

a Obtain the query input information entered by the user;

b, subject and tag analysis are carried out to the query input information to obtain the subject words and tag words corresponding to the query input information;

c, according to the subject word and the tag word, carry out a matching query in the above-mentioned established index to obtain candidate text information that matches the query input information;

d. According to the semantic matching degree between the candidate text information and the query input information, determine the target text information that matches the query input information.

According to another aspect of the present invention, also provides a kind of index establishment equipment for indexing based on text information, wherein, this equipment comprises:

an information determination device, used for determining structured information therefrom according to text information;

a topic extraction device for extracting topic words from the structured information;

A tag determining device is configured to determine a tag word corresponding to the topic from the text information according to the topic corresponding to the topic word;

An index building device is used to build an index for the subject word and the tag word.

According to yet another aspect of the present invention, there is also provided a matching device for matching user's query input information according to the aforementioned established index, wherein the device includes:

a query acquisition device, configured to acquire query input information input by a user;

an information analysis device, configured to perform subject and tag analysis on the query input information to obtain subject words and tag words corresponding to the query input information;

A matching query device, configured to perform a matching query in the index established as described above according to the subject words and tag words, so as to obtain candidate text information that matches the query input information;

A text determining device is configured to determine target text information matching the query input information according to the semantic matching degree between the candidate text information and the query input information.

According to yet another aspect of the present invention, there is also provided a system for building an index and matching query input information of a user, including the aforementioned index building device and the aforementioned matching device.

Compared with the prior art, the present invention determines structured information from text information; extracts subject words from the structured information; Label words corresponding to the topic; build an index for the topic words and the label words. Further, the present invention obtains the query input information input by the user; performs subject and label analysis on the query input information to obtain the subject words and label words corresponding to the query input information; according to the subject words and the label words, A matching query is performed in the aforementioned established index to obtain candidate text information that matches the query input information; matching target text information.

Based on encyclopedic resource knowledge or other resource knowledge mined through the network, the invention extracts the subject and title of the resource knowledge to form an effective description of the content of the resource knowledge, better

This kind of high-quality resource knowledge can be displayed in an efficient manner, which makes the semantic search of this kind of resource knowledge more efficient, meets the complex description search needs that users cannot accurately express with keywords, and improves the user experience.

Description of drawings

Other features, objects and advantages of the present invention will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

1 shows a schematic diagram of a device for indexing based on textual information according to an aspect of the present invention;

2 shows a schematic diagram of a device for establishing an index based on text information according to a preferred embodiment of the present invention;

3 shows a schematic diagram of a device for matching query input information of a user according to another aspect of the present invention;

Figure 4 shows a flowchart of a method for indexing based on textual information according to yet another aspect of the present invention;

Figure 5 shows a flowchart of a method for indexing based on textual information according to yet another aspect of the present invention.

The same or similar reference numbers in the drawings represent the same or similar parts.

Detailed ways

The present invention will be described in further detail below with reference to the accompanying drawings.

Figure 1 shows a schematic diagram of an apparatus for indexing based on textual information according to an aspect of the present invention. The indexing apparatus 1 includes information determining means 101, topic extracting means 102, tag determining means 103, and indexing means 104.

Wherein, the information determination device 101 determines the structured information therefrom according to the text information. Specifically, the information determining device 101 obtains text information, for example, through interaction with data sources, such as encyclopedia data, etc., and further, by structuring the text information, such as analyzing the directory information, subsections contained in the text information Catalog information, etc., from which structured information is determined.

For example, the information determination device 101 obtains the resource knowledge of these encyclopedias as text information through interaction with encyclopedia data such as Baidu Encyclopedia and Interactive Encyclopedia, and further, the information determination device 101 structures the text information, for example, analyzes Catalogs and sub-categories corresponding to each resource knowledge, for example, for the resource knowledge of "disease", the catalog or sub-category corresponding to its symptoms, the catalog or sub-category corresponding to the treatment method, etc. are analyzed.

For another example, the information determination device 101 mines resource knowledge from the Internet by means of data mining as text information, and further structures the text information to determine structured information. For example, the information determining device 101 obtains information such as diseases and disease symptom descriptions, treatment methods, specialized hospitals and other information by mining vertical resource websites. Each resource is organized by disease as an ID. For example, firstly, some candidate seed words are given according to the category, such as disease, coronary heart disease, myocarditis, gastritis, etc., according to the search results, the common top website url is obtained, the structure of the website is analyzed, and the crown is extracted from it. Heart disease, symptoms of coronary heart disease, treatment methods for coronary heart disease, and information on hospitals specializing in coronary heart disease, and incorporate the above information into "diseases" such as coronary heart disease, and organize the coronary heart disease into a business card for storage. Then the "coronary heart disease" can be used as the final text information, and the corresponding information such as "symptoms of coronary heart disease", "treatment methods of coronary heart disease", and "information of hospitals specializing in coronary heart disease" can be used as the text. Information corresponds to structured information.

Those skilled in the art should understand that the above method for determining structured information is only an example, and other existing or possible future methods for determining structured information, if applicable to the present invention, should also be included within the protection scope of the present invention. and is hereby incorporated by reference.

The subject extraction means 102 extracts subject words from the structured information. Specifically, the subject extracting means 102 extracts subject words from the structured information according to the structured information determined by the information determining means 101, such as through subject classifiers, or other predetermined means of extracting subject words.

Here, the purpose of extracting the subject heading is to extract the subject representing the text information from the text information, so as to serve the establishment of a semantic index and the subsequent semantic matching calculation.

Preferably, the index building device 1 further includes a theme training device (not shown), the theme training device obtains training corpus corresponding to the predetermined theme system according to a predetermined theme system; according to the training corpus, the training theme is classified wherein, the topic extraction device 102 extracts the topic words from the structured information according to the topic classifier.

Specifically, the theme training device determines a predetermined theme system. For example, the theme training device determines the search requirements commonly used by network search users according to the statistical results of query sequences input by a large number of network search users, and combines the currently commonly used classification systems, such as encyclopedia, Knowing and other existing systems, determine the subject classification system with certain needs, and use it as a predetermined subject system. Further, the subject training device obtains training corpus corresponding to the predetermined subject system according to the predetermined subject system. For example, if there is a corresponding position identifier "Medical Health? Internal Medicine" in the article, the data is considered to be a disease category training data. Subsequently, the subject training device trains a subject classifier according to the training corpus, for example, through the training corpus, trains an svm classification model to serve as a subject classifier.

Next, the subject extraction means 102 extracts subject words from the structured information according to the subject classifier trained by the subject training means. For example, the topic extraction device 102 inputs the word "coronary heart disease" and its symptoms, treatment methods and other structured information into the topic classifier, so as to obtain the topic as "disease". For another example, for a new encyclopedia business card, the topic extraction device 102 inputs it into the topic classifier, such as the svm classifier, so as to obtain the topic corresponding to the category of the encyclopedic business card.

Preferably, the topic extraction device 102 can also perform synonym expansion on the extracted topic, for example, perform synonym expansion on the topic "disease" to add a synonymous topic "disease".

Those skilled in the art should understand that the above methods for extracting subject words are only examples, and other existing or future methods for extracting subject words, if applicable to the present invention, should also be included within the protection scope of the present invention, and in This is incorporated herein by reference.

The tag determining means 103 determines the tag word corresponding to the topic from the text information according to the topic corresponding to the topic word. Specifically, the label determination device 103 determines the label word corresponding to the theme from the text information according to the theme word extracted by the theme extraction device 102 and the theme corresponding to the theme word, for example, for disease-themed Text information, the tag determining device 103 determines the following tag words corresponding to the topic: palpitation, shortness of breath, chest tightness, diarrhea, vomiting, limb weakness, and the like.

Preferably, the label determination device 103 includes a candidate determination unit (not shown), a central word determination unit (not shown) and a label determination unit (not shown). Specifically, the candidate determination unit determines at least one candidate tag word corresponding to the topic from the text information according to the topic corresponding to the topic word. The page data is counted for unary, binary, and ternary words, and words that appear in more than a certain number of page data are extracted as candidate tag words.

Then, the central word determination unit determines the corresponding central word according to the at least one candidate tag word. Next, the label determination unit determines a label word corresponding to the topic according to the distance between the at least one candidate label word and the central word.

For example, the central word determination unit merges all the candidate label words according to the previously counted label data, and conducts offline statistics on these candidate label words. The collinear frequency in the document. For any two candidate label words, calculate the similarity between them according to the following formula:

Here, PMI(w', w ₁ ) represents the mutual information score between w'w ₁ , which is defined as P(w) represents the probability of the word w being counted.

Then, the central word determination unit determines which fields of the text information need to be analyzed according to the theme, such as the symptom category of the disease, the poem itself and the explanation part, the description part of the character and so on. Further, extract all the words appearing in the candidate tag words and the corresponding synonyms, and then form a center of these words as the center word corresponding to the at least one candidate tag word.

Next, the label determination unit calculates the distance between each of the at least one candidate label word and the central word. For example, if T is used to represent the central word, the distance between the candidate label word and the central word can be obtained by calculating the following formula:

Here, Num(T) represents the number of words included in the central word.

Subsequently, the label determination unit determines a label word corresponding to the topic according to the distance between the at least one candidate label word and the central word, for example, a candidate label word whose distance from the central word is less than a predetermined threshold is regarded as a label word corresponding to the topic corresponding label words.

Preferably, as shown in Figure 3, the label determination unit makes a time series with the distance between the candidate label ranking and the central word. If the slope of the ranking change is greater than the predetermined slope threshold, the subsequent nodes are truncated, as shown in Figure 3. Ranked 5th to 6th.

Here, the slope threshold value is set empirically, for example, by statistic on the overall distribution of the scores.

Those skilled in the art should understand that the above-mentioned ways of determining tag words are only examples, and other existing or future ways of determining tag words, if applicable to the present invention, should also be included within the protection scope of the present invention, and in This is incorporated herein by reference.

More preferably, the central word determination unit performs filtering processing on the at least one candidate tag word according to a predetermined filtering rule to obtain at least one filtered candidate tag word; according to the at least one filtered candidate tag word; candidate tag words, determine the central word; wherein, the predetermined filtering rule is determined based on at least any one of the following:

- the part of speech of the at least one candidate tag word;

- a wording rule for the at least one candidate tag word;

- the co-occurrence ratio of the at least one candidate tag word with the topic.

Specifically, noise may be introduced in the process of counting the candidate tag words. Therefore, it is necessary to filter the candidate tag words. The central word determination unit filters the at least one candidate tag word according to a predetermined filtering rule. to obtain at least one filtered candidate tag word.

For example, the central word determining unit performs filtering processing on the at least one candidate label word according to the part of speech of the at least one candidate label word, for example, performs first word and last word filtering on the at least one candidate label word.

For another example, the central word determining unit filters the at least one candidate label word according to the wording rule of the at least one candidate label word. For example, the first word of the candidate label word cannot be "ba" or "do". , "by", "bi" and other words, the last word cannot be "dang", "to", "de" and other words.

For another example, the central word determination unit performs filtering processing on the at least one candidate tagged word according to the co-occurrence ratio of the at least one candidate tagged word and the subject. In the web title, the co-occurrence ratio of the at least one candidate tag word and the topic is counted, and only those that have co-occurred with the topic are retained, or the candidate tag words whose co-occurrence ratio with the topic is greater than a predetermined threshold are retained.

Preferably, the central word determination unit performs filtering processing on the at least one candidate tag word according to a combination of any two predetermined filtering rules mentioned above, or a comprehensive consideration of all three predetermined filtering rules.

Subsequently, the central word determination unit determines the central word according to the at least one filtered candidate tag word.

Those skilled in the art should understand that the above-mentioned predetermined filtering rules are only examples. If other existing or future predetermined filtering rules are applicable to the present invention, they should also be included within the protection scope of the present invention, and are incorporated herein by reference. Included here.

The index building means 104 builds an index for the subject word and the tag word. Specifically, the index establishing means 104 builds an index for the subject word and the label word according to the subject word extracted by the subject word extraction means 102 and the label word determined by the label determination means 102.

For example, suppose the document corresponding to coronary heart disease is ID1, and the corresponding importance in this document is WC1(x). For example, x can be equal to "disease", "palpitation and shortness of breath", etc. The document corresponding to myocarditis is ID2, and the document corresponding to gastritis is ID3, the document corresponding to the stroke is ID4. The index establishment means 104 establishes a unified inverted index for subject terms and tag terms in the following manner:

Disease - ID1 (WC1(x)), ID2 (WC2(x)), ID3 (WC3(x)), ID4 (WC4(x))

Frustrated Shortness of Breath - ID1 (WC1(x)), ID2 (WC2(x)), ID4 (WC4(x))

Palpitations Shortness of Breath - ID1 (WC1(x)), ID2 (WC2(x)), ID4 (WC4(x))

Vomit - ID3 (WC3(x)), ID4 (WC4(x))

Spit - ID3 (WC3(x)), ID4 (WC4(x))

Preferably, the index building device 1 further includes a normalization device (not shown), the normalization device determines the normalization of the plurality of semantically consistent label words if the label words include a plurality of semantically consistent label words. The normalization result; wherein, the index building device 104 builds an index for the subject word, the tag word and the normalization result.

Specifically, the label words corresponding to the subject word "disease" may include a plurality of label words with consistent semantics, such as "spit" and "nausea and vomiting", which have the same semantics, and the normalization device determines the normalization of the two label words The normalized result is "vomiting"; then, the indexing means 104 indexes the subject word "disease", the tag words "vomiting", "nausea and vomiting" and the normalized result "vomiting".

Those skilled in the art should understand that the above-mentioned method of establishing an index is only an example, and other existing or possible future methods of establishing an index, if applicable to the present invention, should also be included within the protection scope of the present invention, and are hereby referred to as References are included here.

Usually, indexing is to build indexes for keywords. Here, the indexing device 1 also builds indexes for subject words, tag words and their normalized results, so as to achieve better matching between the user's query input information and resource knowledge.

Preferably, the various devices of the index building apparatus 1 work continuously. Specifically, the information determining means 101 determines structured information from the text information; the topic extracting means 102 extracts subject words from the structured information; the label determining means 103 extracts the subject words from the text according to the topics corresponding to the subject words The tag word corresponding to the topic is determined in the information; the index building device 104 creates an index for the topic word and the tag word. Here, those skilled in the art should understand that “continuous” means that each device of the index building apparatus 1 performs the determination of structured information, the extraction of subject words, the determination of label words and The index is established until the index establishment device 1 stops determining the structured information for a long period of time.

Here, the index building device 1 determines structured information therefrom according to the text information; extracts subject words from the structured information; Corresponding label words; build an index for the subject word and the label word. Based on encyclopedia resource knowledge or other resource knowledge mined through the network, the index building device extracts the subject and title of the resource knowledge to form an effective description of the content of the resource knowledge and better display this kind of high-quality resource knowledge. The semantic search of resource-like knowledge is more efficient, meeting the search needs of complex descriptions that users cannot accurately express with keywords, and improving the user experience.

Figure 2 shows a schematic diagram of a device for matching query input information of a user according to another aspect of the present invention. The matching device 2 includes a query acquisition device 201 , an information analysis device 202 , a matching query device 203 and a text determination device 204 .

Wherein, the query obtaining means 201 obtains the query input information input by the user. Specifically, the user inputs query input information through interaction with the user equipment, and the query acquisition means 201 invokes an application programming interface (API) provided by the user equipment, invokes a dynamic page technology such as JSP, ASP, or PHP, or , and obtain the query input information input by the user through other agreed communication methods.

Here, the query input information includes, but is not limited to, the query input information submitted by the user through different input methods such as text input, voice input, and image input.

Those skilled in the art should understand that the above-mentioned ways of obtaining query input information are only examples, and other existing or possible future ways of obtaining query input information, if applicable to the present invention, should also be included within the protection scope of the present invention. and is hereby incorporated by reference.

The information analysis device 202 performs subject and tag analysis on the query input information to obtain subject words and tag words corresponding to the query input information. Specifically, the information analysis device 202 performs subject and label analysis on the query input information acquired by the query acquisition device 201. For example, by inputting the query input information into the subject classifier obtained by the aforementioned training, the corresponding query input information is obtained. The information analysis device 202 performs tag analysis on the query input information input by the user to obtain the corresponding tag words. Here, the method of the information analysis device 202 to analyze the tags of the query input information is the same as or similar to the method of the aforementioned tag determination device 103 to determine the tag words of the text information, so it will not be repeated here, and it is included in the reference method. this.

The matching query means 203 performs a matching query in the index established by the aforementioned index establishment means 104 according to the subject words and tag words, so as to obtain candidate text information that matches the query input information. Specifically, the matching query device 203 performs a matching query in the index established by the aforementioned index establishment device 104 according to the query input information input by the user obtained by the query obtaining device 201, for example, by means of all matching or partial matching, to obtain the matching query. The text information of the subject word corresponding to the query input information, or the text information of the tag word corresponding to the query input information is hit, as the candidate text information matching the query input information.

For example, assuming that the query input information input by the user is "palpitation and shortness of breath", the query acquisition device 201 obtains the query input information "palpitation and shortness of breath" input by the user; the information analysis device 202 performs tag analysis on the query input information, and the obtained tag word is " palpitation and shortness of breath”, the index established by the aforementioned index building device 104 for the tag word “palpitation and shortness of breath” is as follows:

Frustrated Shortness of Breath - ID1 (WC1(x)), ID2 (WC2(x)), ID4 (WC4(x))

Among them, ID1, ID2, and ID4 respectively represent the ID numbers of the text information containing the tag word "palpitation and shortness of breath", and WC1(x), WC2(x), and WC4(x) respectively represent the label word "palpitation and shortness of breath". importance in several textual information.

Then, the matching query device 203 performs a matching query in the index established by the index building device 104 according to the tag word "palpitation and shortness of breath" corresponding to the query input information of the user, and obtains the query input information "palpitation and shortness of breath" according to the above index. The corresponding candidate text information - text information ID1, ID2 and ID4.

Those skilled in the art should understand that the above matching query methods are only examples, and other existing or possible future matching query methods, if applicable to the present invention, should also be included within the protection scope of the present invention, and are hereby referred to as References are included here.

The text determining means 204 determines the target text information matching the query input information according to the semantic matching degree between the candidate text information and the query input information.

Specifically, there is a certain degree of semantic matching between the candidate text information and the query input information, and the semantic matching degree can be obtained by calculation, or further by calculating the matching between the index word set corresponding to the candidate text information and the query input information The matching degree between word sets is obtained. The text determination device 204 determines the target text information matching the query input information according to the semantic matching degree between the candidate text information and the user's query input information, for example, the candidate text information with the highest semantic matching degree is used as the query input information The matching target text information, or the candidate text information whose semantic matching degree is greater than a predetermined matching degree threshold is used as the target text information matching the query input information.

Here, the predetermined matching degree threshold is the semantic matching degree used to judge whether the candidate text information matches the query input information, and its value can be preset and fixed, or can be adjusted according to the actual situation.

Preferably, the text determination device further includes a matching calculation unit (not shown) and a text determination unit (not shown). The matching calculation unit calculates the semantic matching degree between the candidate text information and the query input information; the text determining unit determines the target text information matching the query input information according to the semantic matching degree and a predetermined matching threshold value .

For example, the matching calculation unit calculates the semantic matching degree between the candidate text information and the user's query input information according to the existing matching degree calculation method; when the semantic matching degree is greater than the predetermined matching degree threshold, the text determining unit The candidate text information serves as the target text information that matches the query input information.

Preferably, the text determination device can also determine the target text information corresponding to the query input information according to the index word set corresponding to the candidate text information and the matching word set corresponding to the query input information. Specifically, the candidate text information has a corresponding set of index words. For example, if the subject corresponding to the candidate text information ID1 in the above example is "coronary heart disease", and the corresponding index words include "disease", "palpitation and shortness of breath", then these indexes The index word set formed by the words is the index word set corresponding to the candidate text information ID1. The user's query input information also has a corresponding matching word set. For example, the matching words are obtained by performing word segmentation on the query input information, and then the set composed of the matching words is used as the matching word set corresponding to the query input information. The query input information input by the user is "palpitation, shortness of breath and vomiting". After the matching device 1 performs word segmentation processing on the query input information, the matching words "palpitation, shortness of breath" and "vomiting" are obtained, and the set formed by the two matching words is The matching word set corresponding to the query input information. The text determination device 204 determines the target text information that matches the query input information of the user according to the index word set and the matching word set, for example, will hit the text information corresponding to the index word set with the most matching words in the matching word set , as the target text information matching the query input information; or, the text information corresponding to the index word set whose number of matching words is greater than a predetermined threshold is used as the target text information matching the query input information.

For example, for the candidate text information ID1, ID2 and ID4 in the above example, the index word set corresponding to ID1 includes the index words "disease" and "palpitation and shortness of breath"; the index word set corresponding to ID2 includes the index words "palpitation and shortness of breath", "vomiting" ", "disease"; the index word set corresponding to ID4 includes the index word "palpitation and shortness of breath". Then, for the query input information "palpitation, shortness of breath and vomiting" input by the user, the matching words are "palpitation and shortness of breath" and "vomiting", and the index word set corresponding to ID2 hits the most matching word in the matching word set corresponding to the query input information. The candidate text information ID2 is used as the target text information that most matches the query input information, or, assuming that the predetermined number threshold is 0, the index word set corresponding to the candidate text information ID1, ID2 and ID4 hits the matching word set. If the number of matching words is greater than the predetermined number threshold, the above-mentioned candidate text information ID1, ID2 and ID4 are all used as target text information matching the query input information. When the matching device 2 is provided to the user, it can be sorted according to the importance of the corresponding index words in the candidate text information.

Those skilled in the art should understand that the above methods for determining target text information are only examples, and other existing or future methods for determining target text information, if applicable to the present invention, should also be included within the protection scope of the present invention. and is hereby incorporated by reference.

Preferably, the various devices of the matching device 2 work continuously. Specifically, the query acquisition means 201 acquires the query input information input by the user; the information analysis means 202 performs subject and tag analysis on the query input information to obtain the subject words and tag words corresponding to the query input information; the matching query means 203 According to the subject words and tag words, a matching query is performed in the index established by the aforementioned index establishment device 104 to obtain candidate text information that matches the query input information; the text determination device 204 is based on the candidate text information and The semantic matching degree of the query input information determines the target text information that matches the query input information. Here, those skilled in the art should understand that “continuous” means that each device of the matching device 2 performs the acquisition of query input information, the analysis of topics and tags, and the matching query of candidate text information according to the set or real-time adjustment requirements of the working mode, respectively. The target text information is determined until the matching device 2 stops acquiring the query input information input by the user for a long period of time.

Here, the devices of the index building device 1 and the matching device 2 cooperate with each other to realize the query input information based on the user input, and obtain the corresponding target text information by matching; It can extract the subject and title of the resource knowledge, form an effective description of the resource knowledge content, better display this kind of high-quality resource knowledge, make the semantic search for this kind of resource knowledge more efficient, and satisfy users who cannot accurately use the key The complex description of the word expression to describe the search requirements improves the user experience.

Preferably, the subject word and the tag word can also be regarded as two different domains, corresponding to the subject domain and the tag domain respectively. A matching query is performed in the corresponding aforementioned index to obtain candidate text information matching the query input information.

Specifically, the matching query means 203 adopts the method of domain matching according to the subject words and tag words obtained by the information analysis means 202 through the analysis of the query input information input by the user, respectively in the indexes corresponding to the subject domain and the tag domain. A match query is performed to obtain candidate text information.

Here, the subject domain and the tag domain can be obtained by analyzing the query input information, for example, using the aforementioned subject classifier to analyze the query input information input by the user to obtain the subject category.

Here, the index corresponding to the subject domain and the label domain is the index established by the aforementioned index establishment device 104, and according to the previously established label, the label word extraction is performed on the query input information input by the user, such as for the query input information contained in the query input information. And in the label set, it will be extracted. Then, use the tag word and topic category to pull down the candidates of the document in the corresponding topic and tag unified index, and take the document containing the topic category or tag as the candidate text information corresponding to the query input information, and participate in the subsequent calculation.

Preferably, the matching query device 203 can also consider the corresponding weights of the subject domain and the tag domain, perform a matching query in the corresponding index, comprehensively consider the corresponding weights of the subject domain and the tag domain, and finally obtain candidate text information.

Preferably, the text determination device 204 determines a target index word set in the index word set corresponding to the candidate text information according to the matching words included in the matching word set, wherein the target index word set hits the The most matching words in the matching word set; if the similarity between the target index word set and the matching word set is greater than a predetermined threshold, the text information corresponding to the target index word set is used as the matching word with the query input information. target text information.

Specifically, the text determining device 204 uses the number of matching words in the matching word set according to the index word set corresponding to the candidate text information, and takes the index word set with the largest number of matching words as the target index word set; then, the text determining device 204 calculates The similarity between the target index word set and the matching word set, for example, calculate the similarity between the target index word set and the matching word set, and the similarity between the hit index word and the corresponding matching word, and then simply add or weight the average, etc. method to calculate the similarity between the target index word set and the matching word set, when the similarity is greater than a predetermined threshold, the text determination device takes the text information corresponding to the target index word set as the target matching the query input information text information.

Here, the predetermined threshold is the similarity threshold for judging whether to use the text information corresponding to the target index word set as the target text information according to the similarity between the target index word set and the matching word set. Adjust the actual situation.

Preferably, the matching device 2 further comprises a word set determination device (not shown). The word set determination device performs word segmentation processing on the query input information to obtain word segmentation processed by the word segmentation; A matching word set corresponding to the query input information is obtained, wherein the words included in the matching word set are used as matching words. Subsequently, the matching calculation unit calculates the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information.

Specifically, the word set determination device performs word segmentation processing on the query input information obtained by the query acquisition device 201 to obtain word segmentation processed by word segmentation. Preferably, the word set determination device can also process the word segmentation to obtain word segmentation Perform filtering processing such as removing stop words, and then obtain the final word segmentation; then, according to the obtained word segmentation, the word set determination device combines it with the subject words and tag words obtained by the aforementioned information analysis device 202. After processing, etc., to finally obtain a matching word set corresponding to the query input information, and use the words included in the matching word set as matching words corresponding to the query input information.

Subsequently, the matching calculation unit calculates the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information.

More preferably, the matching device 2 further includes a subsequent processing device (not shown). The subsequent processing device performs subsequent processing on the matching words to update the matching word set; wherein, the subsequent processing includes at least any one of the following:

- determining mutually synonymous matched words included in the matched words, and merging the mutually synonymous matched words into a subset of the set of matched words.

- performing synonym expansion on the matching words, and determining the synonyms obtained after the synonym expansion and the matching words as subsets of the matching word set.

Specifically, the subsequent processing means performs subsequent processing on the matching words in the matching word set determined by the word set determining means, so as to update the matching word set. For example, the subsequent processing device determines mutually synonymous matching words included in the matching words, and combines the mutually synonymous matching words into a subset of the matching word set. Since the matching words may include mutually synonymous matching words, such as "vomit" and "spit", the post-processing device combines these mutually synonymous matching words into a subset of the matching word set.

For example, assuming that the query input information input by the user is Q, the word set determination device performs word segmentation processing on the query input information, and after filtering processing such as removing stop words, the matching word set in the label field is expressed as Q={a,b, c, d, e}, where a, b, c, d, and e are the matching words included in the matching word set; assuming that the matching words a and b are mutually synonymous matching words, the subsequent processing device Combining the matching words a and b into a subset of the matching word set, the matching word set update is expressed as Q={{a,b},c,d,e}. Subsequently, subsequent devices, such as the matching query device 203, perform subsequent matching query operations.

For another example, the subsequent processing device further performs synonym expansion on the matching word, and determines the synonym obtained after the synonym expansion and the matching word as a subset of the matching word set. Specifically, the subsequent processing device may also synonymously expand the matching words in the matching word set corresponding to the query input information, for example, synonymously expand "palpitation and shortness of breath" to "palpitation and shortness of breath", and then, the subsequent processing device will synonymously expand the same word. The synonym obtained after the sense expansion and the matching word are determined as a subset of the matching word set.

Continuing from the above example, for the matching word set Q={{a,b},c,d,e} after synonymous merging, the subsequent processing device can also perform synonym expansion on the matching word set, and expand to obtain the The synonym of the matching word abcde is determined, and the synonym obtained after the synonym expansion and the matching word are determined as subsets of the matching word set. For example, after the matching word set Q is synonymously expanded for many times, the following expression is obtained:

Then, the matching query device 203 performs a matching query in the index established by the index establishment device 104 according to the matching word set, for example, through an inverted index, obtains a candidate text information.

Assuming that the index word set that hits the most matching words in the matching word set is denoted by C, then C is:

where C represents the maximum number of synonymous hits The word set of the position semantic mapping corresponding to w _1i

Then, the matching calculation unit calculates the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information.

The semantic matching degree between Q and C can be calculated by the following formula:

in, express word The weight of , here is represented by (log(TF)+1)*log(N/DF); Match(T _Q ,T _C ) represents whether the index word set, matching word set and topic match.

Here, the value corresponding to Match(T _Q , T _C ) can be defined. If it is assumed that the index word set and the matching word set match the topic, the value of Match (T _Q , T _C ) is 1, otherwise it is 0.5.

Then, assuming that the calculated semantic matching degree value is greater than a predetermined threshold, the text determining unit takes the text information corresponding to the index word set as the target text information matching the query input information.

Figure 4 shows a flowchart of a method for indexing based on textual information according to yet another aspect of the present invention.

In step S401, the indexing apparatus 1 determines structured information therefrom according to the textual information. Specifically, in step S401, the index building device 1 obtains text information, for example, through interaction with a data source, such as encyclopedia data, etc., and further, by structuring the text information, such as analyzing the text information contained in the Directory information, sub-directory information, etc., from which structured information is determined.

For example, in step S401, the index building device 1 obtains the resource knowledge of these encyclopedias as text information by interacting with the encyclopedia data such as Baidu Baike, Interactive Baike, etc., and then, in step S401, the index building device 1 for the The text information is structured, for example, by analyzing the catalogues and subcategories corresponding to each resource knowledge, for example, for the resource knowledge of "disease", analyzing the catalogues or subcategories corresponding to the symptoms, and the catalogues or subcategories corresponding to the treatment methods, etc.

For another example, in step S401, the index building device 1 mines resource knowledge from the Internet by means of data mining as text information, and further structures the text information to determine structured information. For example, in step S401, the index building device 1 obtains information such as diseases and disease symptom descriptions, treatment methods, specialized hospitals and other information by mining vertical resource websites. Each resource is organized by disease as an ID. For example, firstly, some candidate seed words are given according to the category, such as diseases, coronary heart disease, myocarditis, gastritis, etc. are given, and according to the search results, the url of the website with the top ranking is obtained, the structure of the website is analyzed, and the crown is extracted from it. Heart disease, symptoms of coronary heart disease, treatment methods for coronary heart disease, and information on hospitals specializing in coronary heart disease, and incorporate the above information into "diseases" such as coronary heart disease, and organize the coronary heart disease into a business card for storage. Then the "coronary heart disease" can be used as the final text information, and the corresponding information such as "symptoms of coronary heart disease", "treatment methods of coronary heart disease", and "information of hospitals specializing in coronary heart disease" can be used as the text. Information corresponds to structured information.

Those skilled in the art should understand that the above method for determining structured information is only an example, and other existing or possible future methods for determining structured information, if applicable to the present invention, should also be included within the protection scope of the present invention. and is hereby incorporated by reference.

In step S402, the indexing apparatus 1 extracts the subject words from the structured information. Specifically, in step S402, the index building device 1 extracts subject words from the structured information according to the structured information determined in step S401, for example, through a subject classifier, or other predetermined methods of extracting subject words.

Here, the purpose of extracting the subject heading is to extract the subject representing the text information from the text information, so as to serve the establishment of a semantic index and the subsequent semantic matching calculation.

Preferably, the method further includes step S405 (not shown). In step S405, the index building device 1 obtains training corpus corresponding to the predetermined theme system according to the predetermined theme system; according to the training corpus, the training theme classifier; wherein, in step S402, the index building device 1 extracts the subject word from the structured information according to the subject classifier.

Specifically, in step S405, the index building device 1 determines a predetermined subject system. For example, in step S405, the index building device 1 determines the search requirements commonly used by network search users according to the statistical results of query sequences input by a large number of network search users, And combined with the currently commonly used classification systems, such as encyclopedia, knowing and other existing systems, determine the subject classification system with certain needs, and use it as the predetermined subject system. Further, in step S405, the index building device 1 obtains the training corpus corresponding to the predetermined subject system according to the predetermined subject system. Considered the training corpus for disease categories. Subsequently, in step S405, the index building device 1 trains a subject classifier according to the training corpus, for example, through the training corpus, trains an svm classification model as a subject classifier.

Next, in step S402, the index building apparatus 1 extracts subject words from the structured information according to the subject classifier trained in step S405. For example, in step S402, the index building device 1 inputs the word "coronary heart disease" and structured information such as symptoms and treatment methods into the topic classifier, thereby obtaining the topic as "disease". For another example, for a new encyclopedia business card, in step S402, the index building device 1 inputs it into the subject classifier, such as the svm classifier, so as to obtain the subject corresponding to the category of the encyclopedic business card.

Preferably, in step S402, the index building device 1 can also perform synonym expansion on the extracted topic, for example, perform synonym expansion on the topic "disease" to add a synonymous topic "disease".

Those skilled in the art should understand that the above methods for extracting subject words are only examples, and other existing or future methods for extracting subject words, if applicable to the present invention, should also be included within the protection scope of the present invention, and in This is incorporated herein by reference.

In step S403, the index building device 1 determines a tag word corresponding to the topic from the text information according to the topic corresponding to the topic word. Specifically, in step S403, the index building device 1 determines the tag word corresponding to the topic from the text information according to the topic word extracted in step S402 and the topic corresponding to the topic word, for example, for For text information on the subject of disease, in step S403, the indexing device 1 determines the following tag words corresponding to the subject: palpitation, shortness of breath, chest tightness, diarrhea, vomiting, limb weakness, and the like.

Preferably, step S403 further includes sub-step S403a (not shown), sub-step S403b (not shown) and sub-step S403c (not shown). Specifically, in sub-step S403a, the index building device 1 determines at least one candidate tag word corresponding to the topic from the text information according to the topic corresponding to the topic word, for example, in sub-step S403a , the index building device 1 performs unigram, binary and ternary word statistics on all page data organized by vocabulary, and extracts words that appear in page data larger than a certain number as candidate tag words.

Subsequently, in sub-step S403b, the index building device 1 determines the corresponding central word according to the at least one candidate tag word. Next, in sub-step S403c, the index building device 1 determines a tag word corresponding to the topic according to the distance between the at least one candidate tag word and the center word.

For example, in sub-step S403b, the index building device 1 combines all the candidate tag words according to the previously counted tag data, and performs offline statistics on these candidate tag words. The statistical process is as follows: Network-wide data, the collinear frequency of documents in the statistics. For any two candidate label words, calculate the similarity between them according to the following formula:

Here, PMI(w', w ₁ ) represents the mutual information score between w'w ₁ , which is defined as P(w) represents the probability of the word w being counted.

Subsequently, in sub-step S403b, the indexing device 1 determines which fields of the text information need to be analyzed according to the subject, such as the symptom category of the disease, the poem itself and the explanation part, the description part of the character, etc. Further, extract all the words appearing in the candidate label words and the corresponding synonyms, and then form a center of these words as the center word corresponding to the at least one candidate label word.

Next, in sub-step S403c, the index building device 1 calculates the distance between each of the at least one candidate tag word and the center word. For example, if T is used to represent the center word, the distance between the candidate tag word and the center word can be Obtained by the following formula:

Here, Num(T) represents the number of words included in the central word.

Subsequently, in sub-step S403c, the index building device 1 determines a tag word corresponding to the topic according to the distance between the at least one candidate tag word and the center word, for example, selects a candidate whose distance from the center word is less than a predetermined threshold Label words as the label words corresponding to the topic.

Preferably, as shown in FIG. 3, in sub-step S403c, the index building device 1 makes a time series with the distance between the candidate label ranking and the central word. If the slope of the ranking change is greater than the predetermined slope threshold, the subsequent nodes are Cut off, ranking points 5 to 6 in Figure 3.

Here, the slope threshold value is set empirically, for example, by statistic on the overall distribution of the scores.

Those skilled in the art should understand that the above-mentioned ways of determining tag words are only examples, and other existing or future ways of determining tag words, if applicable to the present invention, should also be included within the protection scope of the present invention, and in This is incorporated herein by reference.

More preferably, in sub-step S403b, the index building device 1 performs filtering processing on the at least one candidate tag word according to a predetermined filtering rule to obtain at least one filtered candidate tag word; Filter the processed candidate tag words to determine the central word; wherein, the predetermined filtering rule is determined based on at least any one of the following:

- the part of speech of the at least one candidate tag word;

- a wording rule for the at least one candidate tag word;

- the co-occurrence ratio of the at least one candidate tag word with the topic.

Specifically, noise may be introduced in the process of counting the candidate tag words. Therefore, it is necessary to filter the candidate tag words. In sub-step S403b, the index building device 1, according to a predetermined filtering rule, filters the at least one candidate tag word. The tag words are filtered to obtain at least one filtered candidate tag word.

For example, in sub-step S403b, the index building device 1 performs filtering processing on the at least one candidate tag word according to the part of speech of the at least one candidate tag word, for example, performs first and last word filtering on the at least one candidate tag word.

For another example, in sub-step S403b, the index building device 1 performs filtering processing on the at least one candidate tag word according to the wording rule of the at least one candidate tag word, for example, the first word of the candidate tag word cannot be " ", "do", "by", "bi" and other words, the last word can not be "dang", "to", "de" and other words.

For another example, in sub-step S403b, the index building device 1 performs filtering processing on the at least one candidate label word according to the co-occurrence ratio between the at least one candidate label word and the topic. For example, in sub-step S403b, index building The device 1 counts the co-occurrence ratio between the at least one candidate tag word and the topic in the search statistics log and in the title of the entire network, and only the co-occurrence ratio with the topic is retained, or the co-occurrence ratio with the topic is retained. Candidate tag words greater than a predetermined threshold.

Preferably, in sub-step S403b, the index building device 1 performs filtering processing on the at least one candidate tag word according to a combination of any two predetermined filtering rules mentioned above or a comprehensive consideration of all three predetermined filtering rules.

Subsequently, in sub-step S403b, the index building device 1 determines the central word according to the at least one filtered candidate tag word.

Those skilled in the art should understand that the above-mentioned predetermined filtering rules are only examples. If other existing or future predetermined filtering rules are applicable to the present invention, they should also be included within the protection scope of the present invention, and are incorporated herein by reference. Included here.

In step S404, the index building device 1 builds an index for the subject word and the tag word. Specifically, in step S404, the index building device 1 builds an index for the subject word and the label word according to the subject word extracted in step S402 and the label word determined in step S402.

For example, suppose the document corresponding to coronary heart disease is ID1, and the corresponding importance in this document is WC1(x). For example, x can be equal to "disease", "palpitation and shortness of breath", etc. The document corresponding to myocarditis is ID2, and the document corresponding to gastritis is ID3, the document corresponding to the stroke is ID4. In step S404, the index establishment device 1 establishes a unified inverted index for the subject word and the tag word in the following manner:

Disease - ID1 (WC1(x)), ID2 (WC2(x)), ID3 (WC3(x)), ID4 (WC4(x))

Frustrated Shortness of Breath - ID1 (WC1(x)), ID2 (WC2(x)), ID4 (WC4(x))

Palpitations Shortness of Breath - ID1 (WC1(x)), ID2 (WC2(x)), ID4 (WC4(x))

Vomit - ID3 (WC3(x)), ID4 (WC4(x))

Spit - ID3 (WC3(x)), ID4 (WC4(x))

Preferably, the method further includes step S406 (not shown). In step S406, if the label word includes a plurality of label words with consistent semantics, the index building device 1 determines the classification of the plurality of label words with consistent semantics. The normalization result; wherein, in step S404, the index building device 1 builds an index for the subject word, the label word and the normalization result.

Specifically, the tag word corresponding to the subject word "disease" may include a plurality of tag words with consistent semantics, such as "spit" and "nausea and vomiting", that is, the semantics are the same, then in step S406, the index building device 1 determines the two The normalized result of the tag word is "vomit"; then, in step S404, the index building device 1 creates the subject word "disease", the tag word "vomit", "nausea and vomiting" and the normalized result "vomit" index.

Those skilled in the art should understand that the above-mentioned method of establishing an index is only an example, and other existing or possible future methods of establishing an index, if applicable to the present invention, should also be included within the protection scope of the present invention, and are hereby referred to as References are included here.

Usually, indexing is to build indexes for keywords. Here, the indexing device 1 also builds indexes for subject words, tag words and their normalized results, so as to achieve better matching between the user's query input information and resource knowledge.

Preferably, the various steps of the index building apparatus 1 are continuously working. Specifically, in step S401, the index building device 1 determines structured information from the text information; in step S402, the index building device 1 extracts subject words from the structured information; in step S403, the index building device 1. According to the topic corresponding to the topic word, determine the tag word corresponding to the topic from the text information; in step S404, the index building device 1 creates an index for the topic word and the tag word. Here, those skilled in the art should understand that “continuous” means that each step of the index building device 1 performs the determination of structured information, the extraction of subject words, the determination of label words and The index is established until the index establishment device 1 stops determining the structured information for a long period of time.

Here, the index building device 1 determines structured information therefrom according to the text information; extracts subject words from the structured information; Corresponding label words; build an index for the subject word and the label word. Based on encyclopedia resource knowledge or other resource knowledge mined through the network, the index building device extracts the subject and title of the resource knowledge to form an effective description of the content of the resource knowledge and better display this kind of high-quality resource knowledge. The semantic search of resource-like knowledge is more efficient, meeting the search needs of complex descriptions that users cannot accurately express with keywords, and improving the user experience.

Figure 5 shows a flowchart of a method for indexing based on textual information according to yet another aspect of the present invention.

In step S501, the matching device 2 obtains the query input information input by the user. Specifically, the user inputs the query input information through the interaction with the user equipment. In step S501, the matching device 2 calls the application programming interface (API) provided by the user equipment, and dynamically calls such as JSP, ASP, or PHP. The page technology, or through other agreed communication methods, obtains the query input information input by the user.

Here, the query input information includes, but is not limited to, the query input information submitted by the user through different input methods such as text input, voice input, and image input.

Those skilled in the art should understand that the above-mentioned ways of obtaining query input information are only examples, and other existing or possible future ways of obtaining query input information, if applicable to the present invention, should also be included within the protection scope of the present invention. and is hereby incorporated by reference.

In step S502, the matching device 2 performs subject and tag analysis on the query input information to obtain subject words and tag words corresponding to the query input information. Specifically, in step S502, the matching device 2 performs subject and label analysis on the query input information obtained in step S501, for example, by inputting the query input information into the subject classifier obtained by the aforementioned training, to obtain the query input The subject word corresponding to the information; in step S502, the matching device 2 performs tag analysis on the query input information input by the user to obtain the corresponding tag word. Here, in step S502, the method of the tag analysis of the query input information by the matching device 2 is the same as or similar to the manner in which the aforementioned index establishment device 1 determines the tag word of the text information in step S403, so it is not repeated here. and are incorporated herein by reference.

In step S503, the matching device 2 performs a matching query in the index established by the aforementioned index building device 1 according to the subject words and tag words, so as to obtain candidate text information that matches the query input information. Specifically, in step S503, the matching device 2 performs a matching query in the index established by the aforementioned index establishment device 1 according to the query input information input by the user obtained in step S501, for example, by means of full matching or partial matching, Obtain the text information that hits the subject word corresponding to the query input information, or the text information that hits the tag word corresponding to the query input information, as candidate text information matching the query input information.

For example, assuming that the query input information input by the user is "palpitation and shortness of breath", in step S501, the matching device 2 obtains the query input information "palpitation and shortness of breath" input by the user; in step S502, the matching device 2 labels the query input information After analysis, the obtained tag word is "palpitation and shortness of breath", and the index established by the aforementioned index building device 1 for the label word "palpitation and shortness of breath" is as follows:

Frustrated Shortness of Breath - ID1 (WC1(x)), ID2 (WC2(x)), ID4 (WC4(x))

Among them, ID1, ID2, and ID4 respectively represent the ID numbers of the text information containing the tag word "palpitation and shortness of breath", and WC1(x), WC2(x), and WC4(x) respectively represent the label word "palpitation and shortness of breath". importance in several textual information.

Then in step S503, the matching device 2 performs a matching query in the index established by the index building device 1 according to the tag word "palpitation and shortness of breath" corresponding to the query input information of the user, and obtains the query input information according to the above index. Candidate text information corresponding to "palpitation and shortness of breath" - text information ID1, ID2 and ID4.

Those skilled in the art should understand that the above matching query methods are only examples, and other existing or possible future matching query methods, if applicable to the present invention, should also be included within the protection scope of the present invention, and are hereby referred to as References are included here.

In step S504, the matching device 2 determines the target text information matching the query input information according to the semantic matching degree between the candidate text information and the query input information.

Specifically, there is a certain degree of semantic matching between the candidate text information and the query input information, and the semantic matching degree can be obtained by calculation, or further by calculating the matching between the index word set corresponding to the candidate text information and the query input information The matching degree between word sets is obtained. In step S504, the matching device 2 determines the target text information matching the query input information according to the semantic matching degree between the candidate text information and the query input information of the user, for example, the candidate text information with the highest semantic matching degree is used as the candidate text information with the highest semantic matching degree. The target text information that matches the query input information, or the candidate text information whose semantic matching degree is greater than a predetermined matching degree threshold is used as the target text information matching the query input information.

Here, the predetermined matching degree threshold is the semantic matching degree used to judge whether the candidate text information matches the query input information, and its value can be preset and fixed, or can be adjusted according to the actual situation.

Preferably, this step S504 further includes sub-step S504a (not shown) and sub-step S504b (not shown). In sub-step S504a, the matching device 2 calculates the semantic matching degree between the candidate text information and the query input information; in sub-step S504a, the matching device 2 determines the matching degree according to the semantic matching degree combined with a predetermined matching degree threshold. The query input information matches the target text information.

For example, in sub-step S504a, the matching device 2 calculates the semantic matching degree between the candidate text information and the user's query input information according to the existing matching degree calculation method; when the semantic matching degree is greater than the predetermined matching degree threshold, then in In sub-step S504b, the matching device 2 takes the candidate text information as the target text information matching the query input information.

Preferably, in step S504, the matching device 2 can also determine the target text information corresponding to the query input information according to the index word set corresponding to the candidate text information and the matching word set corresponding to the query input information. Specifically, the candidate text information has a corresponding set of index words. For example, it is assumed that the subject corresponding to the candidate text information ID1 in the above example is "coronary heart disease", and the corresponding index words include "disease", "palpitation and shortness of breath", etc., then these indexes The index word set formed by the words is the index word set corresponding to the candidate text information ID1. The user's query input information also has a corresponding matching word set. For example, the matching words are obtained by performing word segmentation on the query input information, and then the set formed by the matching words is used as the matching word set corresponding to the query input information. The query input information input by the user is "palpitation, shortness of breath and vomiting". After the matching device 1 performs word segmentation processing on the query input information, the matching words "palpitation and shortness of breath" and "vomiting" are obtained, and the set formed by the two matching words is The matching word set corresponding to the query input information. In step S504, the matching device 2 determines the target text information matching the query input information of the user according to the index word set and the matching word set. The corresponding text information is used as the target text information that matches the query input information; or, the text information corresponding to the index word set whose number of hit matching words is greater than the predetermined number threshold is used as the target text information that matches the query input information. text information.

For example, for the candidate text information ID1, ID2 and ID4 in the above example, the index word set corresponding to ID1 includes the index words "disease" and "palpitation and shortness of breath"; the index word set corresponding to ID2 includes the index words "palpitation and shortness of breath", "vomiting" ", "disease"; the index word set corresponding to ID4 includes the index word "palpitation and shortness of breath". Then, for the query input information "palpitation, shortness of breath and vomiting" input by the user, the matching words are "palpitation and shortness of breath" and "vomiting", and the index word set corresponding to ID2 hits the most matching word in the matching word set corresponding to the query input information. The candidate text information ID2 is used as the target text information that most matches the query input information, or, assuming that the predetermined number threshold is 0, the index word set corresponding to the candidate text information ID1, ID2 and ID4 hits the matching word set. If the number of matching words is greater than the predetermined number threshold, the above-mentioned candidate text information ID1, ID2 and ID4 are all used as target text information matching the query input information. When the matching device 2 is provided to the user, it can be sorted according to the importance of the corresponding index words in the candidate text information.

Those skilled in the art should understand that the above methods for determining target text information are only examples, and other existing or future methods for determining target text information, if applicable to the present invention, should also be included within the protection scope of the present invention. and is hereby incorporated by reference.

Preferably, the various steps of the matching device 2 work continuously. Specifically, in step S501, the matching device 2 obtains the query input information input by the user; in step S502, the matching device 2 performs subject and tag analysis on the query input information to obtain the subject corresponding to the query input information word and label word; in step S503, the matching device 2 performs a matching query in the index established by the aforementioned index building device 1 according to the subject word and label word to obtain candidate text information that matches the query input information ; In step S504, the matching device 2 determines the target text information that matches the query input information according to the semantic matching degree between the candidate text information and the query input information. Here, those skilled in the art should understand that "continuous" means that each step of the matching device 2 performs the acquisition of query input information, the analysis of topics and tags, and the matching query of candidate text information according to the set or real-time adjustment of the working mode requirements, respectively. The target text information is determined until the matching device 2 stops acquiring the query input information input by the user for a long period of time.

Here, the steps of the index building device 1 and the matching device 2 cooperate with each other to realize the query input information based on the user input, and obtain the corresponding target text information by matching; It can extract the subject and title of the resource knowledge, form an effective description of the resource knowledge content, better display this kind of high-quality resource knowledge, make the semantic search for this kind of resource knowledge more efficient, and satisfy users who cannot accurately use the key The complex description of the word expression to describe the search requirements improves the user experience.

Preferably, the subject word and the tag word can also be regarded as two different domains, corresponding to the subject domain and the tag domain respectively. In step S503, the matching device 2, according to the subject word and the tag word, respectively, in the subject domain and the tag word A matching query is performed in the aforementioned index corresponding to the domain to obtain candidate text information that matches the query input information.

Specifically, in step S503, the matching device 2 adopts the method of sub-domain matching according to the subject words and tag words obtained by analyzing the query input information input by the user in step S502, respectively in the subject field and the tag field. A matching query is performed in the corresponding index to obtain candidate text information.

Here, the subject domain and the tag domain can be obtained by analyzing the query input information, for example, using the aforementioned subject classifier to analyze the query input information input by the user to obtain the subject category.

Here, the index corresponding to the subject domain and the label domain is the index established by the aforementioned index establishment device 1, and according to the previously established label, the label word extraction is performed on the query input information input by the user, such as for the query input information contained in the query input information. And if it is in the tag set, it will be extracted. Then, use the tag word and topic category to pull down the candidates of the document in the corresponding topic and tag unified index, and take the document containing the topic category or tag as the candidate text information corresponding to the query input information, and participate in the subsequent calculation.

Preferably, in step S503, the matching device 2 may also consider the weights corresponding to the subject domain and the tag domain, perform a matching query in the corresponding index, comprehensively consider the corresponding weights of the subject domain and the tag domain, and finally obtain a candidate. text information.

Preferably, in step S504, the matching device 2 determines a target index word set in the index word set corresponding to the candidate text information according to the matching words included in the matching word set, wherein the target index word set hits The most matching words in the matching word set; if the similarity between the target index word set and the matching word set is greater than a predetermined threshold, the text information corresponding to the target index word set is used as the query input information. Matching target text information.

Specifically, in step S504, the matching device 2 hits the number of matching words in the matching word set according to the index word set corresponding to the candidate text information, and takes the index word set with the largest number of hit matching words as the target index word set; then, in step S504 In S504, the matching device 2 calculates the similarity between the target index word set and the matching word set, for example, calculates the similarity between the target index word set and the matching word set, the hit index word and the corresponding matching word respectively, and then passes Simple addition or weighted average, etc., calculate the similarity between the target index word set and the matching word set, when the similarity is greater than a predetermined threshold, in step S504, the matching device 2 The text corresponding to the target index word set information as the target text information that matches the query input information.

Here, the predetermined threshold is the similarity threshold for judging whether to use the text information corresponding to the target index word set as the target text information according to the similarity between the target index word set and the matching word set. Adjust the actual situation.

Preferably, the method further includes step S505 (not shown). In step S505, the matching device 2 performs word segmentation processing on the query input information to obtain the word segmentation processed by the word segmentation; the word segmentation and the subject words and tag words obtained by the matching device 2 in step S502 A merging process is performed to obtain a matching word set corresponding to the query input information, wherein the words included in the matching word set are used as matching words. Subsequently, in sub-step S504a, the matching device 2 calculates the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information.

Specifically, in step S505, the matching device 2 performs word segmentation processing on the query input information obtained in step S501 to obtain the word segmentation processed by the word segmentation. Preferably, in step S505, the matching device 2 can also perform word segmentation on After the word segmentation is processed, the word segmentation is obtained, and filtering processing such as removing stop words is performed to obtain the final word segmentation; then, in step S505, the matching device 2 compares the obtained word segmentation with the word segmentation obtained by the matching device 2 in step S502. Subject words and tag words are combined, de-redundant, etc., to finally obtain a matching word set corresponding to the query input information, and use the words included in the matching word set as the matching words corresponding to the query input information. .

Subsequently, in sub-step S504a, the matching device 2 calculates the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information.

More preferably, the method further includes step S506 (not shown). In step S506, the matching device 2 performs subsequent processing on the matching words to update the matching word set; wherein, the subsequent processing includes at least any one of the following:

- determining mutually synonymous matched words included in the matched words, and merging the mutually synonymous matched words into a subset of the set of matched words.

- performing synonym expansion on the matching words, and determining the synonyms obtained after the synonym expansion and the matching words as subsets of the matching word set.

Specifically, in step S506, the matching device 2 performs subsequent processing on the matching words in the matching word set determined in step S505 to update the matching word set. For example, in step S506, the matching device 2 determines mutually synonymous matching words included in the matching words, and combines the mutually synonymous matching words into a subset of the matching word set. Since the matching words may include mutually synonymous matching words, such as "vomit" and "spit", in step S506, the matching device 2 combines these mutually synonymous matching words into a subset of the matching word set.

For example, assuming that the query input information input by the user is Q, in step S505, the matching device 2 performs word segmentation processing on the query input information, and after filtering processing such as removing stop words, the matching word set in the tag field is expressed as Q={ a,b,c,d,e}, where a,b,c,d,e are the matching words included in the matching word set; assuming that the matching words a and b are mutually synonymous matching words, Then in step S506, the matching device 2 combines the matching words a and b into a subset of the matching word set, and the matching word set is updated and expressed as Q={{a,b},c,d,e}. Subsequently, subsequent steps such as step S503 perform subsequent matching query operations.

For another example, in step S506, the matching device 2 further performs synonym expansion on the matching word, and determines the synonym and the matching word obtained after the synonym expansion as a subset of the matching word set. Specifically, in step S506, the matching device 2 can also synonymously expand the matching words in the matching word set corresponding to the query input information, for example, synonymously expand "palpitation and shortness of breath" to "palpitation and shortness of breath", then, in step S506 In S506, the matching device 2 determines the synonym obtained after the synonym expansion and the matching word as a subset of the matching word set.

Continuing from the previous example, for the matched word set Q={{a,b},c,d,e} after synonymous merging, in step S506, the matching device 2 can also synonymously expand the matched word set, Expand to obtain the synonyms of the matching words a, b, c, d, and e, and determine the synonym and the matching word obtained after the synonym expansion as a subset of the matching word set, for example, the matching word set Q After multiple synonym expansions, the following expression is obtained:

Subsequently, in step S503, the matching device 2 performs a matching query in the index established by the index building device 1 according to the matching word set. candidate text information.

Assuming that the index word set that hits the most matching words in the matching word set is denoted by C, then C is:

where C represents the maximum number of synonymous hits The word set of the position semantic mapping corresponding to w _1i

Then in sub-step S504a, the matching device 2 calculates the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information.

The semantic matching degree between Q and C can be calculated by the following formula:

in, express word The weight of , here is represented by (log(TF)+1)*log(N/DF); Match(T _Q ,T _C ) represents whether the index word set, matching word set and topic match.

Here, the value corresponding to Match(T _Q , T _C ) can be defined. If it is assumed that the index word set and the matching word set match the topic, the value of Match (T _Q , T _C ) is 1, otherwise it is 0.5.

Then, assuming that the calculated semantic matching degree value is greater than a predetermined threshold, in sub-step S504b, the matching device 2 takes the text information corresponding to the index word set as the target text information matching the query input information.

It should be noted that the present invention may be implemented in software and/or a combination of software and hardware, for example, may be implemented using an application specific integrated circuit (ASIC), a general purpose computer, or any other similar hardware device. In one embodiment, the software program of the present invention may be executed by a processor to implement the steps or functions described above. Likewise, the software programs of the present invention (including associated data structures) may be stored on computer-readable recording media, such as RAM memory, magnetic or optical drives, or floppy disks, and the like. Additionally, some of the steps or functions of the present invention may be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

In addition, a part of the present invention can be applied as a computer program product, such as computer program instructions, which when executed by a computer, through the operation of the computer, can invoke or provide methods and/or technical solutions according to the present invention. Rather, program instructions for invoking the method of the present invention may be stored in fixed or removable recording media, and/or transmitted via data streams in broadcast or other signal-bearing media, and/or stored in accordance with the in the working memory of the computer device on which the program instructions are executed. Here, an embodiment according to the present invention includes an apparatus including a memory for storing computer program instructions and a processor for executing the program instructions, wherein, when the computer program instructions are executed by the processor, a trigger is The apparatus operates based on the aforementioned methods and/or technical solutions according to various embodiments of the present invention.

It will be apparent to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, but that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments are to be regarded in all respects as illustrative and not restrictive, and the scope of the invention is to be defined by the appended claims rather than the foregoing description, which are therefore intended to fall within the scope of the claims. All changes within the meaning and range of the equivalents of , are included in the present invention. Any reference signs in the claims shall not be construed as limiting the involved claim. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Several units or means recited in the device claims can also be realized by one unit or means by means of software or hardware. The terms first, second, etc. are used to denote names and do not denote any particular order.

Claims (17)

1. A method for indexing based on textual information, wherein the method comprises the following steps: A analyze the directory information and/or sub-directory information contained in the text information, structure the acquired text information, and determine the structured information therefrom; B extracts subject headings from the structured information; C1 determines at least one candidate tag word corresponding to the topic from the text information according to the topic corresponding to the topic word; C2 determines the corresponding central word according to the at least one candidate label word; C3 determines the tag word corresponding to the topic according to the distance between the at least one candidate tag word and the center word; D builds an index for the subject term and the tag term. 2. The method of claim 1, wherein the method further comprises: -Acquiring training corpus corresponding to the predetermined theme system according to the predetermined theme system; - Train a topic classifier according to the training corpus; Wherein, the step B includes: - extracting the subject words from the structured information according to the subject classifier. 3. The method according to claim 1, wherein the step C2 comprises: - according to a predetermined filtering rule, filtering the at least one candidate tag word to obtain at least one filtered candidate tag word; - determining the central word according to the at least one filtered candidate tag word; Wherein, the predetermined filtering rule is determined based on at least any one of the following: - the part of speech of the at least one candidate tag word; - a wording rule for the at least one candidate tag word; - the co-occurrence ratio of the at least one candidate tag word with the topic. 4. The method of any one of claims 1 to 3, wherein the method further comprises: - if the label word includes a plurality of semantically consistent label words, determining the normalization result of the plurality of semantically consistent label words; Wherein, the step D includes: - Indexing the subject term, the tag term and the normalized result. 5. A method for matching user's query input information based on the index established according to claim 1, wherein the method comprises the following steps: a Obtain the query input information entered by the user; b Perform subject and tag analysis on the query input information to obtain subject words and tag words corresponding to the query input information; c, carry out a matching query in the index established as claimed in claim 1 according to the subject words and tag words, to obtain candidate text information that matches the query input information; d. According to the semantic matching degree between the candidate text information and the query input information, determine the target text information that matches the query input information. 6. The method of claim 5, wherein the step d comprises: d1 calculates the semantic matching degree between the candidate text information and the query input information; d2, according to the semantic matching degree, combined with a predetermined matching degree threshold, to determine the target text information that matches the query input information. 7. The method of claim 6, wherein the method further comprises: - Perform word segmentation processing on the query input information to obtain the word segmentation processed by the word segmentation; - Combine the segmented words with the subject words and tag words obtained in step b to obtain a matching word set corresponding to the query input information, wherein the words included in the matching word set are used as matching words; Wherein, the step d1 includes: - Calculate the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information. 8. The method of claim 7, wherein the method further comprises: - performing subsequent processing on the matching words to update the matching word set; Wherein, the subsequent processing includes at least any one of the following: - determining mutually synonymous matching words included in the matching words, and merging the mutually synonymous matching words into a subset of the matching word set; - Performing synonym expansion on the matching words, and determining the synonyms obtained after the synonym expansion and the matching words as subsets of the matching word set. 9. An indexing device for indexing based on textual information, wherein the device comprises: an information determination device, configured to analyze the directory information and/or sub-directory information contained in the text information, structure the acquired text information, and determine the structured information therefrom; a topic extraction device for extracting topic words from the structured information; Labeling devices, including: a candidate determination unit, configured to determine at least one candidate tag word corresponding to the topic from the text information according to the topic corresponding to the topic word; a central word determination unit, configured to determine a corresponding central word according to the at least one candidate tag word; A label determination unit, configured to determine a label word corresponding to the topic according to the distance between the at least one candidate label word and the center word; An index building device is used to build an index for the subject word and the tag word. 10. The indexing device according to claim 9, wherein the device further comprises subject training means for: -Acquiring training corpus corresponding to the predetermined theme system according to the predetermined theme system; - Train a topic classifier according to the training corpus; Wherein, the subject extraction device is used for: - extracting the subject words from the structured information according to the subject classifier. 11. The indexing device according to claim 9, wherein the central word determining unit is used for: - according to a predetermined filtering rule, filtering the at least one candidate tag word to obtain at least one filtered candidate tag word; - determining the central word according to the at least one filtered candidate tag word; Wherein, the predetermined filtering rule is determined based on at least any one of the following: - the part of speech of the at least one candidate tag word; - a wording rule for the at least one candidate tag word; - the co-occurrence ratio of the at least one candidate tag word with the topic. 12. The indexing device according to any one of claims 9 to 11, wherein the device further comprises: a normalization device, configured to determine a normalization result of the plurality of semantically consistent label words if the label word includes a plurality of semantically consistent label words; Wherein, the index establishment device is used for: - Indexing the subject term, the tag term and the normalized result. 13. A matching device for matching user's query input information based on the index established according to claim 9, wherein the device comprises: a query acquisition device, configured to acquire query input information input by a user; an information analysis device, configured to perform subject and tag analysis on the query input information to obtain subject words and tag words corresponding to the query input information; A matching query device, configured to perform a matching query in the index established as claimed in claim 10 according to the subject words and tag words, to obtain candidate text information that matches the query input information; A text determination device is configured to determine target text information matching the query input information according to the semantic matching degree between the candidate text information and the query input information. 14. The apparatus of claim 13, wherein the text determining means comprises: a matching calculation unit, configured to calculate the semantic matching degree between the candidate text information and the query input information; A text determination unit, configured to determine target text information that matches the query input information according to the semantic matching degree in combination with a predetermined matching degree threshold. 15. The apparatus according to claim 14, wherein the apparatus further comprises word set determining means for: - Perform word segmentation processing on the query input information to obtain the word segmentation processed by the word segmentation; - Combine the segmented words with the subject words and tag words obtained by the information analysis device to obtain a matching word set corresponding to the query input information, wherein the words included in the matching word set are used as matching words word; Wherein, the matching calculation unit is used for: - Calculate the semantic matching degree between the candidate text information and the query input information according to the matching word set and the index word set corresponding to the candidate text information. 16. The apparatus according to claim 15, wherein the apparatus further comprises post-processing means for: - performing subsequent processing on the matching words to update the matching word set; Wherein, the subsequent processing includes at least any one of the following: - determining mutually synonymous matching words included in the matching words, and merging the mutually synonymous matching words into a subset of the matching word set; - Performing synonym expansion on the matching words, and determining the synonyms obtained after the synonym expansion and the matching words as subsets of the matching word set. 17. A system for indexing and matching user's query input information, comprising the indexing device as described in any one of claims 9 to 12, and the system as described in any one of claims 13 to 16 match the device.