CN108763555A - Representation data acquisition methods and device based on demand word - Google Patents

Abstract

The invention discloses a method and device for acquiring portrait data based on demand words. Among them, the method includes: establishing a knowledge map and/or an event map in a specific field; establishing a tag map based on user behavior data; obtaining input demand words belonging to a specific field; performing searches on the knowledge map and/or event map respectively, and obtaining the corresponding The knowledge sub-graph and/or the matter sub-graph corresponding to the demand word; the label sub-graph corresponding to the knowledge sub-graph and/or the matter sub-graph is retrieved in the label map; the portrait data matching the demand word is obtained according to the label sub-graph. This solution can quickly and effectively generate portrait data that matches the input demand words through the knowledge map and/or event map, as well as the label map, avoiding the low extraction efficiency and high labor cost caused by manual extraction of user portraits in the prior art Moreover, the acquired portrait data can fully and accurately reflect the user portrait corresponding to the demand word.

Description

Image data acquisition method and device based on demand words

technical field

The invention relates to the technical field of data processing, in particular to a method and device for acquiring portrait data based on demand words.

Background technique

User portrait is a user model abstracted by classifying or labeling user information. For a certain product or demand, its user portrait data is the comprehensive tagged data of the target user of the product or demand. For example, the user profile data for a certain game product may be: male, loves to play games, and unmarried.

At present, in the process of obtaining portrait data corresponding to specific needs, the method of manual construction is usually adopted, using the knowledge and experience of personnel in one or several specific fields, and based on user data closely related to the specific needs, to construct and User portraits corresponding to specific needs. However, this method is inefficient in obtaining portrait data corresponding to specific needs, and the labor cost is high. The result of the selected portrait data is highly dependent on the builder, which is not conducive to large-scale application; and, in view of the limitation of the builder's knowledge domain, and Due to the limitations of the user data closely related to the specific needs, the final profile data obtained is often one-sided and cannot truly reflect the user profile corresponding to the specific needs.

Contents of the invention

In view of the above problems, the present invention is proposed to provide a method and device for acquiring portrait data based on demand words to overcome the above problems or at least partially solve the above problems.

According to one aspect of the present invention, a method for acquiring portrait data based on demand words is provided, which includes:

Build domain-specific knowledge graphs and/or event graphs;

Based on user behavior data, create a tag map;

Obtain the input demand words belonging to a specific field;

Retrieve respectively on the knowledge map and/or the matter map to obtain the knowledge sub-graph and/or matter sub-graph corresponding to the demand words;

Retrieving a tag sub-graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph;

According to the label sub-graph, the portrait data matching the demand word is obtained.

According to another aspect of the present invention, a kind of portrait data acquisition device based on demand words is provided, which includes:

The first building module is suitable for building a knowledge graph and/or an event graph in a specific field;

The second building module is adapted to create a tag map according to user behavior data;

An acquisition module, adapted to acquire input demand words belonging to a specific field;

The first retrieval module is adapted to perform retrieval on the knowledge graph and/or the affair graph respectively, to obtain the knowledge sub-graph and/or the affair sub-graph corresponding to the demand word;

The second retrieval module is adapted to retrieve a tag sub-graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph;

The portrait acquisition module is adapted to obtain portrait data matching the demand word according to the tag sub-map.

According to yet another aspect of the present invention, a computing device is provided, including: a processor, a memory, a communication interface, and a communication bus, and the processor, the memory, and the communication interface complete mutual communication through the communication bus communication;

The memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the method for acquiring portrait data based on demand words.

According to another aspect of the present invention, a computer storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction causes the processor to execute the method corresponding to the above-mentioned demand word-based portrait data acquisition method operate.

According to the method and device for obtaining portrait data based on demand words provided by the present invention. Among them, first establish a knowledge map and/or an event map in a specific field; and establish a tag map based on user behavior data; secondly, obtain the input demand words belonging to a specific field, and search them separately in the knowledge map and/or event map to obtain The knowledge sub-graph and/or the matter sub-graph corresponding to the demand word, and finally retrieve the label sub-graph corresponding to the knowledge sub-graph and/or the matter sub-graph in the label map, and obtain the portrait data matching the demand word according to the label sub-graph . This solution can quickly and effectively generate portrait data that matches the input demand words through the knowledge map and/or event map, as well as the label map, avoiding the low extraction efficiency and high labor cost caused by manual extraction of user portraits in the prior art Moreover, the use of knowledge graphs and/or event graphs in this scheme can broaden the knowledge domain of data in a specific field, make full use of the correlation between things or events, and obtain knowledge sub-graphs and/or event graphs corresponding to demand words. Li sub-graph, and obtain portrait data according to knowledge sub-graph and/or event sub-graph, so that the finally obtained portrait data can more comprehensively and accurately reflect the user portrait corresponding to the demand word.

The above description is only an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the description, and in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable , the specific embodiments of the present invention are enumerated below.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiment. The drawings are only for the purpose of illustrating a preferred embodiment and are not to be considered as limiting the invention. Also throughout the drawings, the same reference numerals are used to designate the same components. In the attached picture:

Fig. 1 shows a schematic flow diagram of a method for acquiring portrait data based on demand words according to an embodiment of the present invention;

Fig. 2 shows a schematic flow diagram of a method for acquiring portrait data based on demand words according to another embodiment of the present invention;

Fig. 3a shows a schematic diagram of a knowledge map provided according to another embodiment of the present invention;

Fig. 3b shows a schematic diagram of an event map provided according to another embodiment of the present invention;

Figure 3c shows a schematic diagram of a label map provided according to another embodiment of the present invention;

Fig. 4 shows the structural block diagram of the portrait data acquisition device based on demand words provided according to one embodiment of the present invention;

Fig. 5 shows a schematic structural diagram of a computing device provided according to an embodiment of the present invention.

Detailed ways

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided for more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

Fig. 1 shows a schematic flowchart of a method for acquiring portrait data based on demand words according to an embodiment of the present invention. As shown in Figure 1, the method includes:

Step S110, establishing a domain-specific knowledge map and/or event map.

Among them, the knowledge map is a structured semantic knowledge base, which is used to describe the concepts and their relationships in the physical world. Its basic unit is the triplet of "knowledge entity-relationship-knowledge entity". Connections form a web-like knowledge structure. The event map is a logical knowledge base that describes the development, causality and other relationships of events with the triplet of "event entity-relationship-event entity" as the core.

In the process of obtaining the portrait data corresponding to the demand word, the demand word often corresponds to a specific field, such as the financial field, the social field, or the catering field and so on. Therefore, in this step, it is first necessary to establish a domain-specific knowledge map and/or event map.

The methods for constructing knowledge graphs and event graphs can be set by those skilled in the art according to actual business requirements, which are not limited in this embodiment. For example, data in a specific field may be obtained, and information extraction and integration may be performed on the data in the specific field to form a knowledge map and/or an event map in the specific field.

In step S120, a tag graph is established according to user behavior data.

Different from the knowledge map and event map described in step S110, the tag map established in this step is not based on data in a specific field, but is based on user behavior data, after extracting tagged information from user behavior data, and obtaining the The association relationship, construct the "label entity-relationship-label entity" triplet, so as to establish the label map.

Step S130, acquiring the input demand words belonging to a specific field.

Wherein, the demand words in this step are not limited to directly input demand words in a specific field, but also include demand words belonging to a specific field extracted from the input demand information. Moreover, the present invention does not limit the specific input method, for example, it may be a combination of one or more input methods such as manual input, voice input, and body recognition input.

In step S140, search is performed on the knowledge graph and/or the event graph respectively to obtain the knowledge sub-graph and/or the event sub-graph corresponding to the demand word.

Specifically, the knowledge entity matching the demand word can be retrieved in the knowledge graph, and the knowledge sub-graph corresponding to the demand word can be determined according to the knowledge entity; event entity, and determine the event sub-graph corresponding to the demand word according to the event entity.

Through this step, the knowledge sub-graph and/or event sub-graph associated with the demand word can be obtained, avoiding the disadvantages of low comprehensiveness and accuracy of the portrait data obtained directly from the demand word in the tag map.

Step S150, searching the tag graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph.

According to the knowledge sub-graph and/or the matter sub-graph obtained in step S140, the label sub-graph corresponding to the knowledge sub-graph and/or the matter sub-graph is retrieved through a corresponding retrieval method. For example, according to the similarity or correlation between each knowledge entity or event entity in the knowledge sub-graph and/or event sub-graph and the label entity in the label graph, the label corresponding to the knowledge sub-graph and/or event sub-graph can be determined For the sub-graph, this embodiment does not limit the specific method of retrieving the label sub-graph corresponding to the knowledge sub-graph and/or the affairs sub-graph, and those skilled in the art can set it according to actual business needs.

Step S160, according to the label sub-map, obtain the portrait data matching the required words.

According to the tag sub-map determined in step S150, the portrait data matching the required word is obtained through tag extraction or semantic analysis and other methods.

In this embodiment, a knowledge map and/or an event map of a specific field is first established, and a label map is established according to user behavior data; secondly, the input demand words belonging to a specific field are obtained, and searches are performed on the knowledge map and/or event map respectively , get the knowledge sub-graph and/or matter sub-graph corresponding to the demand word, and finally retrieve the label sub-graph corresponding to the knowledge sub-graph and/or matter sub-graph in the label map, and obtain the matching demand word according to the label sub-graph portrait data. Therefore, based on the knowledge map and/or the event map, as well as the label map, the portrait data matching the input demand words can be quickly and effectively generated, avoiding the low extraction efficiency and high labor cost caused by manual extraction of user portraits in the prior art Moreover, the use of knowledge graphs and/or event graphs in this scheme can broaden the knowledge domain of data in a specific field, make full use of the correlation between things or events, and obtain knowledge sub-graphs and/or event graphs corresponding to demand words. Li sub-graph, and obtain portrait data according to knowledge sub-graph and/or event sub-graph, so that the finally obtained portrait data can more comprehensively and accurately reflect the user portrait corresponding to the demand word.

Fig. 2 shows a schematic flowchart of a method for acquiring portrait data based on demand words according to another embodiment of the present invention. As shown in Figure 2, the method includes:

Step S210, establishing a domain-specific knowledge map and/or event map.

Specifically, in the process of establishing the knowledge map of a specific field, the initial knowledge map of the specific field can be constructed first: Among them, in the process of building the initial knowledge map, it is first necessary to set each knowledge entity and the corresponding relationship type in the specific field , such as setting the knowledge entity "bank", the knowledge entity "Bank of China" and the corresponding relationship between the two as the relationship between inclusion and inclusion, and then according to the crawled public data (including structured knowledge graph triple data and non- Structured text data), artificially extract the knowledge entities and the relationship between knowledge entities in the unstructured text data, thus forming knowledge graph triple data, and according to the structured knowledge graph triple data and unstructured The knowledge map triplet data generated by the simplified text data is integrated into the initial knowledge map; further, after the initial knowledge map is constructed, the initial knowledge map is used as a training set to construct the entity extraction model and entity relationship classification model of the knowledge map, Among them, the specific method of constructing the entity extraction model of the knowledge graph and the entity relationship classification model is not limited in the present invention. For example, the BiLSTM+CRF model can be used to construct the entity extraction model of the knowledge graph, and the knowledge graph can be constructed according to the BiGRU+2attention model. Entity-relationship classification model; further, after constructing the entity extraction model and entity-relationship classification model of the knowledge map, crawl the structured data and unstructured text data of specific domain knowledge classes, and use the constructed entity extraction model and entity-relationship The classification model extracts knowledge map triples from the structured data and unstructured text data, expands the initial knowledge map, and obtains the constructed knowledge map. For example, Figure 3a shows part of the constructed knowledge graph in the financial field. It can be seen from Figure 3a that the nodes in the knowledge graph are knowledge entities, and the knowledge entities are the corresponding knowledge entity relationships, such as knowledge The entities "savings", "exchange" and "loan" are all business attributes of the knowledge entity "bank"; while the knowledge entity "installment" belongs to the category of the knowledge entity "loan".

In the same way, in the process of establishing an event map in a specific field, an initial event map in a specific field can be constructed, and the initial event map can be used as a training set to build an entity extraction model and an entity relationship classification model for the event map; Structured data and unstructured text data, according to the entity extraction model and entity relationship classification model of the event map, extract the triplet of the event map from the structured data and unstructured text data, expand the initial event map, and obtain the constructed Affair map. Figure 3b is a part of the graph of affairs in the constructed financial field. From Figure 3b, it can be seen that the event "marriage" will lead to the occurrence of the event "buy a house" and/or "travel", and the event "buy a house" will lead to the event " Loans" occur.

In step S220, a tag graph is established according to user behavior data.

Different from the knowledge map and event map in step S210, the label map established in this step is not based on specific field data, but by obtaining user behavior data and extracting user label data from user behavior data, according to users and user labels The association relationship between data establishes a tag graph, wherein the relationship between tag entities in the tag graph is obtained according to the association relationship between users and user tag data. Optionally, after the user tag data is extracted, corresponding descriptive tags can be further obtained according to the user tag data, and a tag map can be established according to the correspondence between the descriptive tags and user tags and the association between users and user tag data. As shown in Figure 3c, through the statistical analysis of user behavior data, it can be concluded that 90% of the users have installed both the Anjuke app and the Autohome app, and 80% of the users who have installed the Autohome app have installed Ant Huabei , establish the relationship between the tags "Anjuke" and "Car Home", and establish the relationship between the tags "Car Home" and "Ant Huabei", and, for the tags "Anjuke" and "Fang Tianxia " can obtain the corresponding description tag "buying a house", then further establish the relationship between the label "buying a house" and the tags "Anjuke" and "Fangtianxia"; similarly, establish the corresponding Describe the association relationship of the label "buy a car", and establish the corresponding relationship between "Guaziwang", "Ant Huabei", and "Jingdong Baitiao" and the corresponding description label "Stage". It can be seen from the figure that the tag map constructed in this step does not record user information (such as user ID, user name, etc.), but presents the tag data obtained according to user behavior data into the tag map, thereby avoiding the The addition of user information to the tag map will increase the amount of data processing and reduce the processing speed.

Step S230, acquiring the input demand words belonging to a specific field.

Wherein, the demand words in this step are not limited to directly input demand words in a specific field, but also include demand words belonging to a specific field extracted from the input demand information through semantic analysis and other methods. Moreover, the present invention does not limit the specific input method, for example, it may be a combination of one or more input methods such as manual input, voice input, and body recognition input.

Step S240, searching for at least one generalized word corresponding to the required word.

Specifically, the underlying corpus is pre-built. Wherein, the specific construction method of the underlying corpus can be set by those skilled in the art, which is not limited in this embodiment. For example, the original corpus of the knowledge graph, the event graph, and/or the tag graph can be used as the corpus for constructing the bottom corpus, so as to construct the bottom corpus. Optionally, to facilitate the search efficiency of subsequent generalized words, the corpus data in the underlying corpus can be converted into corresponding word vectors. For example, data cleaning can be performed on the corpus data, and the word segmentation operation can be performed on the cleaned corpus data, and word vectors can be generated from word segmentation results through a word vector training model (such as the CBOW word2vec word vector training model), and the underlying corpus containing word vectors can be obtained.

Further, at least one generalized word corresponding to the required word is searched in the underlying corpus. In the specific implementation process, the demand word can be analyzed in advance to generate the corresponding demand word vector, and according to the distance between the demand word vector and the word vectors of each corpus in the underlying corpus, at least one corresponding to the demand word can be found generalized words. Specifically, the distance between the required word vector and the word vectors of each corpus in the underlying corpus can be sorted from small to large, and the first n word vectors in the sorted position column can be obtained, and the words corresponding to the n word vectors Determined as a generalized word corresponding to the demand word. For example, if the demand word is "loan", the 4 word vectors closest to the word vector of "loan" can be obtained through the underlying corpus, and the words "borrowing", "borrowing money" and "borrowing" corresponding to the 4 word vectors , and "loan" as a generalization corresponding to "loan".

Step S250, search on the knowledge map to obtain the knowledge sub-graph corresponding to the demand word and at least one generalized word; and/or search on the affair graph to obtain the affair sub-graph corresponding to the demand word and at least one generalized word .

Specifically, in the process of searching the knowledge map and obtaining the knowledge sub-graph corresponding to the demand word and at least one generalized word, at least one knowledge entity ( For example, the knowledge entity corresponding to the demand word and at least one generalization word can be determined according to the Euclidean distance between the knowledge entity in the knowledge map and the demand word or generalization word), and according to the at least one knowledge entity, through the corresponding knowledge Graph relationship prediction model to obtain the corresponding knowledge sub-graph. Among them, the knowledge graph relationship prediction model can be based on a machine learning method, using triple data in the knowledge graph as a training set, and constructed through the RPEM (representation prediction embedding model) model. For example, in the knowledge graph shown in Figure 3a, the knowledge graph sub-graph corresponding to a generalized word "loan" of the demand word is retrieved as a knowledge graph sub-graph containing the knowledge entities "staging", "bank" and "private loan" .

Similarly, in the process of retrieving the event map and obtaining the event sub-map corresponding to the demand word and at least one generalized word, at least one event entity ( For example, the event entity corresponding to the demand word and at least one generalization word can be determined according to the size of the Euclidean distance between the event entity in the event map and the demand word or generalization word), and according to the at least one event entity, through the corresponding event Graph relationship prediction model to obtain the corresponding event sub-graph. Among them, the relationship prediction model of the affair map can be based on the machine learning method, using the triple data in the matter map as the training set, and constructed through the RPEM model. For example, in the event graph shown in Figure 3b, the event graph sub-graph corresponding to the demand word "loan" is retrieved as the event entity "buying a house".

Step S260, searching the tag graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph.

Specifically, when retrieving the tag sub-graph corresponding to the knowledge sub-graph in the tag graph, the tags corresponding to each knowledge entity in the knowledge sub-graph in the tag graph can be determined first, and the corresponding tags associated with the tag graph can be determined according to the corresponding tag graph relationship prediction model. The label sub-map of . Among them, the label map relationship prediction model can be based on machine learning methods, using the "label entity-relationship-label entity" triplet data in the label map as the training set, and constructed through the RPEM model. For example, if the knowledge graph sub-graph determined in step S250 contains the knowledge entity "Stage", then it is determined that the tag entity "Stage" in the tag graph shown in Figure 3c corresponds to it, and it is further determined that the tag "Ant Huabei" is included , "Jingdong Baitiao" and/or "Guaziwang" tag sub-graph is the tag sub-graph corresponding to the knowledge sub-graph.

Similarly, when retrieving the tag sub-map corresponding to the event sub-map in the tag map, you can first determine the tags corresponding to each event entity of the event sub-map in the tag map, and determine the relationship with the tag according to the corresponding tag map relationship prediction model The label sub-map of . For example, if the event entity “buying a house” is included in the event entity graph sub-graph determined in step S250, then it is determined that the tag entity “buying a house” in the tag graph shown in FIG. The label sub-graph of "Fang Tianxia" and/or "Car Home" is the label sub-graph corresponding to this matter sub-graph.

Step S270, according to the label sub-map, obtain the portrait data matching the demand word.

Specifically, according to each tag entity in the tag sub-map obtained in step S260, the portrait data matching the demand word is determined. For example, if it is determined in step S260 that the label sub-maps containing the labels "Ant Huabei", "Jingdong Baitiao", "Guazi.com", "Anjuke", "Fangtianxia" and/or "Car Home" are the corresponding labels For the sub-map, the portrait data matching the demand word includes the use and/or download of Ant Huabei, Jingdong Baitiao, Guazi.com, Anjuke, Fangtianxia and/or "Autohome" application portrait data.

In an optional implementation manner, after obtaining the portrait data matching the demand words, further mining the user groups matching the demand words according to the acquired portrait data matching the demand words. It is used to recommend corresponding products or services for the user group. For example, if the demand word is "loan", loan applications can be recommended for the mined user group, thereby improving the promotion efficiency and effect. Optionally, after the user group matching the demand word is mined, the user group can be further assigned a label corresponding to the demand word, so as to facilitate rapid retrieval of corresponding users according to the label.

In another optional implementation manner, non-high-quality users can be filtered according to the acquired portrait data matching the demand words. For example, in the financial field, it is usually necessary to match the corresponding loan amount or the use authority of financial products according to the user's credit value. In order to reduce the risk value of financial products, it is often necessary to screen out high-risk users of financial products. The method provided in this embodiment can obtain the user portrait corresponding to "low credit" only by inputting the demand word "low credit", so as to set the loan limit or reduce the use of financial products for the user corresponding to the user portrait authority.

In yet another optional implementation, after obtaining the portrait data matching the demand word, the promotion effect of the product or service corresponding to the demand word is estimated according to the proportion of users corresponding to the portrait data in all users, For example, if the profile data matching the demand word "game" includes "men" and "college students", and the current male college students account for 20% of the total population, then estimate the products or services corresponding to the "demand word" based on this ratio or, after obtaining the portrait data matching the demand word, adjust the promotion channel of the product or service corresponding to the demand word, following the above example, according to the user portrait data "male" corresponding to the demand word "game" and "college students", determine the network promotion channel as the main promotion channel for the product or service corresponding to the demand word "game".

In yet another optional implementation manner, attribution processing of portrait data may be implemented according to this embodiment. Specifically, the target portrait data is obtained, and the portrait data corresponding to the plurality of demand words are respectively obtained through the method provided in this embodiment, and the similarity comparison between the target portrait data and the portrait data corresponding to the plurality of demand words is carried out. Determine at least one demand word corresponding to the target portrait data. For example, if the target portrait data is "male" and "college student", and the portrait data corresponding to the demand word "game" is "male" and "college student", the portrait data corresponding to the demand word "cosmetics" is "female " and "white-collar", and the portrait data corresponding to the demand word "health care products" is "middle-aged and elderly", then it can be determined that the demand word corresponding to the target portrait data is "game", so as to match the corresponding products for known user groups or the technical effects of the Services.

In this embodiment, through the established knowledge map and/or event map of a specific field, and the tag map established according to user behavior data, the portrait data matching the demand word is determined, which avoids the need to manually extract user portraits in the prior art. The disadvantages of low extraction efficiency and high labor costs are caused; and, in this solution, by obtaining the generalization words of the demand words, and determining the final portrait data according to the demand words and generalization words, the comprehensiveness of the obtained user portrait data can be improved and accuracy; further, the use of knowledge graphs and/or event graphs in this scheme can broaden the knowledge domain of data in a specific field, make full use of the correlation between things or events, and obtain knowledge sub-graphs and/or graphs corresponding to demand words or matter sub-graph, and obtain portrait data according to the knowledge sub-graph and/or matter sub-graph, so that the final portrait data can more comprehensively and accurately reflect the user portrait corresponding to the demand word, in order to tap new users and avoid risky users , to estimate the promotion effect of products or services, adjust promotion strategies, and provide a basis for attribution processing of target portrait data.

Fig. 4 shows a structural block diagram of a device for acquiring portrait data based on demand words according to an embodiment of the present invention. As shown in FIG. 4 , the device includes: a first establishment module 41 , a second establishment module 42 , an acquisition module 43 , a first retrieval module 44 , a second retrieval module 45 , and a portrait acquisition module 46 .

Among them, the first building module 41 is suitable for building a knowledge map and/or an event map of a specific field.

The second building module 42 is adapted to create a tag map according to user behavior data.

The obtaining module 43 is adapted to obtain the input demand words belonging to a specific field.

The first retrieval module 44 is adapted to perform retrieval on the knowledge graph and/or the event graph respectively, to obtain the knowledge sub-graph and/or the event sub-graph corresponding to the demand words.

The second retrieval module 45 is adapted to retrieve a tag sub-graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph.

The portrait acquisition module 46 is adapted to obtain portrait data matching the demand word according to the tag sub-map.

Optionally, the device further includes: a generalization module (not shown in the figure), adapted for the acquisition module to search for at least one generalization corresponding to the demand word after acquiring the input demand word belonging to a specific field word.

The first retrieval module 44 is further adapted to: perform a search on the knowledge graph to obtain a knowledge sub-graph corresponding to the demand word and the at least one generalized word; and/or perform a search on the affair graph to obtain An event sub-graph corresponding to the demand word and the at least one generalized word.

Optionally, the generalization module is further adapted to: search for at least one generalized word corresponding to the required word in the underlying corpus.

Optionally, the generalization module is further adapted to: analyze the required words to generate the required word vectors; according to the distance between the required word vectors and the word vectors of each corpus in the underlying corpus, search and At least one generalized word corresponding to the demand word.

Optionally, the first building module 41 is further adapted to: build a knowledge map according to the following steps:

Constructing an initial knowledge map of a specific field, using the initial knowledge map as a training set, and constructing an entity extraction model and an entity relationship classification model of the knowledge map;

Crawl structured data and unstructured text data of domain-specific knowledge;

Extract knowledge graph triples from the structured data and unstructured text data according to the entity extraction model and entity relationship classification model of the knowledge graph, expand the initial knowledge graph, and obtain the constructed knowledge graph.

Optionally, the first establishment module 41 is further adapted to: construct an event map according to the following steps:

Constructing an initial event map of a specific field, using the initial event map as a training set, and constructing an entity extraction model and an entity relationship classification model of the event map;

Crawl structured data and unstructured text data of domain-specific affairs;

According to the entity extraction model and the entity relationship classification model of the event map, the triplet of the event map is extracted from the structured data and the unstructured text data, and the initial event map is expanded to obtain the constructed event map.

Optionally, the second building module 42 is further adapted to: extract the user label data in the user behavior data; establish a label map according to the association relationship between the user and the user label data, wherein the tag entity in the label map The relationship between is obtained according to the association relationship between users and user label data.

Optionally, the device further includes: a mining module (not shown in the figure), adapted to mine user groups matching demand words according to the portrait data.

Optionally, the device further includes: a tag assignment module, adapted to configure all users in the user group with the requirement words after the mining module mines the user groups matching the demand words according to the portrait data. The label corresponding to the word.

It can be seen that the device establishes a tag graph based on user behavior data; secondly, obtains the input demand words belonging to a specific field, and performs searches on the knowledge graph and/or event graph respectively, and obtains the knowledge sub-graph and/or corresponding to the demand words or matter sub-graph, and finally retrieve the label sub-graph corresponding to the knowledge sub-graph and/or matter sub-graph in the label map, and obtain the portrait data matching the demand word according to the label sub-graph. Therefore, according to the knowledge map and/or the event map, as well as the label map, the portrait data matching the input demand words can be quickly and effectively generated, avoiding the low extraction efficiency and high labor cost caused by manual extraction of user portraits in the prior art Moreover, the use of knowledge graphs and/or event graphs in this device can broaden the knowledge domain of data in a specific field, make full use of the correlation between things or events, and obtain knowledge sub-graphs and/or event graphs corresponding to demand words. Li sub-graph, and obtain portrait data according to knowledge sub-graph and/or event sub-graph, so that the finally obtained portrait data can more comprehensively and accurately reflect the user portrait corresponding to the demand word.

According to one embodiment of the present invention, a non-volatile computer storage medium is provided, the computer storage medium stores at least one executable instruction, and the computer executable instruction can execute the method in any of the above method embodiments.

FIG. 5 shows a schematic structural diagram of a computing device according to an embodiment of the present invention, and the specific embodiment of the present invention does not limit the specific implementation of the computing device.

As shown in FIG. 5 , the terminal may include: a processor (processor) 502 , a communication interface (Communications Interface) 504 , a memory (memory) 506 , and a communication bus 508 .

in:

The processor 502 , the communication interface 504 , and the memory 506 communicate with each other through the communication bus 508 .

The communication interface 504 is configured to communicate with network elements of other devices such as clients or other servers.

The processor 502 is configured to execute the program 510, and may specifically execute relevant steps in the foregoing method embodiments.

Specifically, the program 510 may include program codes including computer operation instructions.

The processor 502 may be a central processing unit CPU, or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computing device may be of the same type, such as one or more CPUs, or may be different types of processors, such as one or more CPUs and one or more ASICs.

The memory 506 is used for storing the program 510 . The memory 506 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

The program 510 can specifically be used to make the processor 502 perform the following operations:

Build domain-specific knowledge graphs and/or event graphs;

Based on user behavior data, create a tag map;

Obtain the input demand words belonging to a specific field;

Retrieve respectively on the knowledge map and/or the matter map to obtain the knowledge sub-graph and/or matter sub-graph corresponding to the demand words;

Retrieving a tag sub-graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph;

According to the label sub-graph, the portrait data matching the demand word is obtained.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

Find at least one generalized word corresponding to the required word;

Searching in the knowledge graph to obtain a knowledge sub-graph corresponding to the demand word and the at least one generalized word;

And/or, perform a search on the event map to obtain an event sub-map corresponding to the demand word and the at least one generalized word.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

At least one generalized word corresponding to the required word is searched in the underlying corpus.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

Analyzing the demand words to generate the demand word vectors;

Searching for at least one generalized word corresponding to the required word according to the distance between the required word vector and the word vectors of each corpus in the underlying corpus.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

Follow the steps below to build a knowledge graph:

Constructing an initial knowledge map of a specific field, using the initial knowledge map as a training set, and constructing an entity extraction model and an entity relationship classification model of the knowledge map;

Crawl structured data and unstructured text data of domain-specific knowledge;

Extract knowledge graph triples from the structured data and unstructured text data according to the entity extraction model and entity relationship classification model of the knowledge graph, expand the initial knowledge graph, and obtain the constructed knowledge graph.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

Follow the steps below to build an Affair Graph:

Constructing an initial event map of a specific field, using the initial event map as a training set, and constructing an entity extraction model and an entity relationship classification model of the event map;

Crawl structured data and unstructured text data of domain-specific affairs;

According to the entity extraction model and the entity relationship classification model of the event map, the triplet of the event map is extracted from the structured data and the unstructured text data, and the initial event map is expanded to obtain the constructed event map.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

Obtain user behavior data;

extracting user tag data in the user behavior data;

A tag graph is established according to the association relationship between the user and the user tag data, wherein the relationship between tag entities in the tag graph is obtained according to the association relationship between the user and the user tag data.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

According to the portrait data mining, the user groups matching the demand words are obtained.

In an optional implementation manner, the program 510 may be specifically configured to enable the processor 502 to perform the following operations:

Configure labels corresponding to the demand words for all users in the user group.

The algorithms and displays presented herein are not inherently related to any particular computer, virtual system, or other device. Various generic systems can also be used with the teachings based on this. The structure required to construct such a system is apparent from the above description. Furthermore, the present invention is not specific to any particular programming language. It should be understood that various programming languages can be used to implement the content of the present invention described herein, and the above description of specific languages is for disclosing the best mode of the present invention.

In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure the understanding of this description.

Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, in order to streamline this disclosure and to facilitate an understanding of one or more of the various inventive aspects, various features of the invention are sometimes grouped together in a single embodiment, figure, or its description. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description are hereby expressly incorporated into this Detailed Description, with each claim standing on its own as a separate embodiment of this invention.

Those skilled in the art can understand that the modules in the device in the embodiment can be adaptively changed and arranged in one or more devices different from the embodiment. Modules or units or components in the embodiments may be combined into one module or unit or component, and furthermore may be divided into a plurality of sub-modules or sub-units or sub-assemblies. All features disclosed in this specification (including accompanying claims, abstract and drawings) and any method or method so disclosed may be used in any combination, except that at least some of such features and/or processes or units are mutually exclusive. All processes or units of equipment are combined. Each feature disclosed in this specification (including accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

Furthermore, those skilled in the art will understand that although some embodiments described herein include some features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention. and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

The various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all functions of some or all components in the device according to the embodiments of the present invention. The present invention can also be implemented as an apparatus or an apparatus program (for example, a computer program and a computer program product) for performing a part or all of the methods described herein. Such a program for realizing the present invention may be stored on a computer-readable medium, or may be in the form of one or more signals. Such a signal may be downloaded from an Internet site, or provided on a carrier signal, or provided in any other form.

It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means can be embodied by one and the same item of hardware. The use of the words first, second, and third, etc. does not indicate any order. These words can be interpreted as names.

The invention discloses: A1. A method for obtaining portrait data based on demand words, which includes:

Build domain-specific knowledge graphs and/or event graphs;

Based on user behavior data, create a tag map;

Obtain the input demand words belonging to a specific field;

Retrieve respectively on the knowledge map and/or the matter map to obtain the knowledge sub-graph and/or matter sub-graph corresponding to the demand words;

Retrieving a tag sub-graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph;

According to the label sub-graph, the portrait data matching the demand word is obtained.

A2. The method according to A1, wherein, after acquiring the input demand words belonging to a specific field, the method further includes: searching for at least one generalized word corresponding to the demand words;

The searching in the knowledge graph and/or the event graph respectively to obtain the knowledge sub-graph and/or the event sub-graph corresponding to the demand word further includes: searching in the knowledge graph to obtain the knowledge sub-graph corresponding to the demand word A knowledge subgraph corresponding to the at least one generalized word;

And/or, performing the retrieval on the knowledge graph and/or the event graph respectively to obtain the knowledge sub-graph and/or the event sub-graph corresponding to the demand word further includes: performing a search on the event graph to obtain the corresponding An event sub-graph corresponding to the requirement word and the at least one generalized word.

A3. The method according to A2, wherein the searching for at least one generalized word corresponding to the demand word further comprises:

At least one generalized word corresponding to the required word is searched in the underlying corpus.

A4. The method according to A3, wherein said searching for at least one generalized word corresponding to said demand word in the underlying corpus further comprises:

Analyzing the demand words to generate the demand word vectors;

Searching for at least one generalized word corresponding to the required word according to the distance between the required word vector and the word vectors of each corpus in the underlying corpus.

A5. The method according to any one of A1-A4, wherein the establishment of a domain-specific knowledge map and/or event map further comprises:

Follow the steps below to build a knowledge graph:

Constructing an initial knowledge map of a specific field, using the initial knowledge map as a training set, and constructing an entity extraction model and an entity relationship classification model of the knowledge map;

Crawl structured data and unstructured text data of domain-specific knowledge;

Extract knowledge graph triples from the structured data and unstructured text data according to the entity extraction model and entity relationship classification model of the knowledge graph, expand the initial knowledge graph, and obtain the constructed knowledge graph.

A6. The method according to any one of A1-A4, wherein the establishment of a domain-specific knowledge map and/or event map further includes:

Follow the steps below to build an Affair Graph:

Constructing an initial event map of a specific field, using the initial event map as a training set, and constructing an entity extraction model and an entity relationship classification model of the event map;

Crawl structured data and unstructured text data of domain-specific affairs;

According to the entity extraction model and the entity relationship classification model of the event map, the triplet of the event map is extracted from the structured data and the unstructured text data, and the initial event map is expanded to obtain the constructed event map.

A7. The method according to any one of A1-A6, wherein said establishing a tag map further comprises:

Obtain user behavior data;

extracting user tag data in the user behavior data;

A tag graph is established according to the association relationship between the user and the user tag data, wherein the relationship between tag entities in the tag graph is obtained according to the association relationship between the user and the user tag data.

A8. The method according to A1, wherein, after obtaining the portrait data matching the demand words according to the label sub-map, the method further includes:

According to the portrait data mining, the user groups matching the demand words are obtained.

A9. The method according to A8, wherein, after the user groups matching the demand words are obtained according to the portrait data mining, the method further includes: configuring all users in the user groups with the requirements The label corresponding to the word.

The present invention also discloses: B10. A device for obtaining portrait data based on demand words, which includes:

The first building module is suitable for building a knowledge graph and/or an event graph in a specific field;

The second building module is adapted to create a tag map according to user behavior data;

An acquisition module, adapted to acquire input demand words belonging to a specific field;

The first retrieval module is adapted to perform retrieval on the knowledge graph and/or the affair graph respectively, to obtain the knowledge sub-graph and/or the affair sub-graph corresponding to the demand word;

The second retrieval module is adapted to retrieve a tag sub-graph corresponding to the knowledge sub-graph and/or the affair sub-graph in the tag graph;

The portrait acquisition module is adapted to obtain portrait data matching the demand word according to the tag sub-map.

B11. The device according to B10, wherein the device further comprises: a generalization module, adapted for the acquisition module to search for at least one generalization corresponding to the demand word after obtaining the input demand word belonging to a specific field Transformation;

The first retrieval module is further adapted to: perform retrieval on the knowledge graph to obtain a knowledge sub-graph corresponding to the demand word and the at least one generalized word; and/or, perform retrieval on the affair graph, An event sub-graph corresponding to the demand word and the at least one generalization word is obtained.

B12. The device according to B11, wherein the generalization module is further adapted to:

At least one generalized word corresponding to the required word is searched in the underlying corpus.

B13. The device according to B12, wherein the generalization module is further adapted to: analyze the demand word to generate the demand word vector;

Searching for at least one generalized word corresponding to the required word according to the distance between the required word vector and the word vectors of each corpus in the underlying corpus.

B14. The apparatus according to any one of B10-B13, wherein the first building block is further adapted to:

Follow the steps below to build a knowledge graph:

Constructing an initial knowledge map of a specific field, using the initial knowledge map as a training set, and constructing an entity extraction model and an entity relationship classification model of the knowledge map;

Crawl structured data and unstructured text data of domain-specific knowledge;

Extract knowledge graph triples from the structured data and unstructured text data according to the entity extraction model and entity relationship classification model of the knowledge graph, expand the initial knowledge graph, and obtain the constructed knowledge graph.

B15. The apparatus according to any one of B10-B13, wherein the first building block is further adapted to:

Follow the steps below to build an Affair Graph:

Constructing an initial event map of a specific field, using the initial event map as a training set, and constructing an entity extraction model and an entity relationship classification model of the event map;

Crawl structured data and unstructured text data of domain-specific affairs;

According to the entity extraction model and the entity relationship classification model of the event map, the triplet of the event map is extracted from the structured data and the unstructured text data, and the initial event map is expanded to obtain the constructed event map.

B16. The apparatus according to any one of B10-B15, wherein the second building block is further adapted to:

extracting user tag data in the user behavior data;

A tag graph is established according to the association relationship between the user and the user tag data, wherein the relationship between tag entities in the tag graph is obtained according to the association relationship between the user and the user tag data.

B17. The device according to B10, wherein the device further comprises:

The mining module is adapted to mine user groups matching demand words according to the portrait data.

B18. The device according to B17, wherein the device further comprises:

The label allocation module is adapted to assign tags corresponding to the demand words to all users in the user group after the mining module mines the user groups matching the demand words according to the portrait data.

The present invention also discloses: C19. A computing device, comprising: a processor, a memory, a communication interface, and a communication bus, and the processor, the memory, and the communication interface complete mutual communication through the communication bus;

The memory is used to store at least one executable instruction, and the executable instruction causes the processor to execute the operation corresponding to the method for acquiring portrait data based on demand words as described in any one of A1-A9.

The present invention also discloses: D20. A computer storage medium, at least one executable instruction is stored in the storage medium, and the executable instruction causes the processor to execute the demand-based word as described in any one of A1-A9. The operation corresponding to the portrait data acquisition method.

Claims (10)

1. a kind of representation data acquisition methods based on demand word comprising：

Establish the knowledge mapping and/or reason collection of illustrative plates of specific area；

According to user behavior data, label collection of illustrative plates is established；

Obtain the demand word for belonging to specific area of input；

It is retrieved respectively in the knowledge mapping and/or reason collection of illustrative plates, obtains knowledge subgraph spectrum corresponding with the demand word And/or reason subgraph spectrum；

Retrieval label subgraph corresponding with knowledge subgraph spectrum and/or reason subgraph spectrum is composed in the label collection of illustrative plates；

It composes to obtain and the matched representation data of demand word according to the label subgraph.

2. according to the method described in claim 1, wherein, it is described obtain input belong to the demand word of specific area after, The method further includes：Search at least one extensive word corresponding with the demand word；

It is described to be retrieved respectively in the knowledge mapping and/or reason collection of illustrative plates, obtain knowledge corresponding with the demand word Collection of illustrative plates and/or reason subgraph spectrum further comprise：Retrieved in the knowledge mapping, obtain with the demand word and it is described extremely The corresponding knowledge subgraph spectrum of a few extensive word；

And/or it is described retrieved respectively in the knowledge mapping and/or reason collection of illustrative plates, obtain corresponding with the demand word Knowledge subgraph is composed and/or reason subgraph spectrum further comprises：Retrieved in the reason collection of illustrative plates, obtain with the demand word and The corresponding reason subgraph spectrum of at least one extensive word.

3. according to the method described in claim 2, wherein, it is described search at least one extensive word corresponding with the demand word into One step includes：

At least one extensive word corresponding with the demand word is searched in bottom corpus.

4. according to the method described in claim 3, wherein, the lookup in bottom corpus is corresponding extremely with the demand word A few extensive word further comprises：

The demand word is parsed, the demand term vector is generated；

According to the distance between the term vector of each language material in the demand term vector and the bottom corpus, search with it is described The corresponding at least one extensive word of demand word.

5. according to the described method of any one of claim 1-4, wherein the knowledge mapping for establishing specific area and/or Reason collection of illustrative plates further comprises：

Knowledge mapping is built according to the following steps：

The initial knowledge collection of illustrative plates of structure specific area builds the reality of knowledge mapping using the initial knowledge collection of illustrative plates as training set Body extraction model and entity relationship disaggregated model；

Crawl the structural data and unstructured text data of domain-specific knowledge class；

According to the entity extraction model of knowledge mapping and entity relationship disaggregated model in the structural data and non-structural text Knowledge mapping triple is extracted in data, initial knowledge collection of illustrative plates is expanded, the knowledge mapping after being built.

6. according to the described method of any one of claim 1-4, wherein the knowledge mapping for establishing specific area and/or Reason collection of illustrative plates further comprises：

Reason collection of illustrative plates is built according to the following steps：

The initial reason collection of illustrative plates for building specific area builds the reality of reason collection of illustrative plates using the initial reason collection of illustrative plates as training set Body extraction model and entity relationship disaggregated model；

Crawl the structural data and unstructured text data of specific area reason class；

According to the entity extraction model of reason collection of illustrative plates and entity relationship disaggregated model in the structural data and non-structural text Reason collection of illustrative plates triple is extracted in data, initial reason collection of illustrative plates is expanded, the reason collection of illustrative plates after being built.

7. according to the method described in any one of claim 1-6, wherein it is described according to user behavior data, establish label figure Spectrum further comprises：

Obtain user behavior data；

Extract the user tag data in the user behavior data；

According to the incidence relation between user and user tag data, label collection of illustrative plates is established, wherein label in the label collection of illustrative plates Relationship between entity is obtained according to the incidence relation between user and user tag data.

8. a kind of representation data acquisition device based on demand word comprising：

First establishes module, is adapted to set up the knowledge mapping and/or reason collection of illustrative plates of specific area；

Second establishes module, is suitable for, according to user behavior data, establishing label collection of illustrative plates；

Acquisition module, suitable for obtaining the demand word for belonging to specific area of input；

First retrieval module obtains and the demand word suitable for being retrieved respectively in the knowledge mapping and/or reason collection of illustrative plates Corresponding knowledge subgraph spectrum and/or reason subgraph spectrum；

It is corresponding with knowledge subgraph spectrum and/or reason subgraph spectrum to be suitable for the retrieval in the label collection of illustrative plates for second retrieval module Label subgraph spectrum；

Portrait acquisition module, suitable for composing to obtain and the matched representation data of demand word according to the label subgraph.

9. a kind of computing device, including：Processor, memory, communication interface and communication bus, the processor, the storage Device and the communication interface complete mutual communication by the communication bus；

The memory makes the processor execute as right is wanted for storing an at least executable instruction, the executable instruction Ask the corresponding operation of representation data acquisition methods based on demand word described in any one of 1-7.

10. a kind of computer storage media, an at least executable instruction, the executable instruction are stored in the storage medium Processor is set to execute the corresponding behaviour of representation data acquisition methods based on demand word as described in any one of claim 1-7 Make.