CN102364469B - A kind of method and device that illustrative sentence retrieval result is ranked up - Google Patents

Abstract

The present invention provides a method and device for sorting example sentence retrieval results, wherein the method includes: A. obtaining the user's query words; B. retrieving matching example sentences containing the query words from the sentence database; C. calculating each matching example sentence The collocation strength between the matching example sentence and the query word is determined by the collocation probability between each query word and the collocation between each query word and other words in the matching example sentence except each query word D. sort each matching example sentence according to the collocation strength between the matching example sentence and the query word. Through the above method, the purpose and demand of the user for language learning can be better met, the user's browsing efficiency can be improved, and the number of times the system responds to meet the user's demand can be reduced at the same time.

Description

A method and device for sorting example sentence retrieval results

【Technical field】

The invention relates to the technical field of natural language processing, in particular to a method and device for sorting example sentence retrieval results.

【Background technique】

With the in-depth development of computer and Internet technology, it is possible for people to use the powerful computing power of computers to obtain the information they need in language learning. The example sentence retrieval system is a powerful tool to help language learners obtain relevant information. Retrieve example sentences that match user input in a large-scale sentence database to help users obtain the correct usage of the relevant language.

However, the existing example sentence retrieval system does not consider the relationship between the query word input by the user in a specific example sentence and the context of the example sentence in the process of sorting the retrieval results, so that the example sentences that are ranked in the forefront of the retrieval results are likely to appear , not the example sentences the user really wants to get.

For example, for the query word input by the user: "improvement" + "efficiency", the following two matching example sentences are obtained:

1. In a sense, the increase in productivity can achieve higher efficiency.

2. This article explains in detail how to improve the efficiency of large-scale retrieval systems.

Generally speaking, when the user enters multiple query words, there are connections between the multiple query words, and what the user wants to see is how these query words are used in combination in the example sentence. In example sentence 2, "improvement" and "efficiency" just form a collocation relationship, which has a strong internal connection. In example sentence 1, "improvement" actually forms a collocation with "productivity", and "improvement" and "efficiency" The connection between is not strong. For the user, it is obvious that example sentence 2 is what he really wants to obtain. Since the existing technology sorts the retrieval results of example sentences, the two cases of example sentence 1 and example sentence 2 cannot be distinguished, so that the retrieval results that are not sufficiently relevant to the user's needs are ranked in the forefront, thereby affecting the user's browsing efficiency and increasing system response times.

【Content of invention】

The technical problem to be solved by the present invention is to provide a method and device for sorting example sentence retrieval results, so as to solve the defects in the existing example sentence retrieval system that affect the browsing efficiency of users and increase the number of system responses.

The technical solution adopted by the present invention to solve the technical problem is to provide a method for sorting example sentence retrieval results, including: A. obtaining the query words of the user; B. retrieving matching example sentences containing the query words from the sentence library; C .Calculate the collocation strength between each matching example sentence and the query word, wherein the collocation strength between the matching example sentence and the query word is divided by the collocation probability between each query word and each query word and the matching example sentence except each query word The collocation probability between other words other than words is determined, and the collocation probability between words refers to the possibility of forming a collocation relationship between words; D. each matching example sentence is carried out according to the collocation strength between the matching example sentence and the query word Sort.

According to a preferred embodiment of the present invention, the sentence database includes a single sentence database or a double sentence database.

According to a preferred embodiment of the present invention, the collocation strength between the matching example sentences and the query words is equal to: the maximum value of the collocation probabilities between each query word and the probability of each query word and the matching example sentences except each query word The ratio of the maximum value among the collocation probabilities between other words, or the maximum value among the collocation probabilities between each query word and the collocation probability between each query word and other words in the matching example sentence except each query word or the ratio of the average of the collocation probabilities between each query word to the average of the collocation probabilities between each query word and words other than each query word in the matching example sentence, or , the difference between the average of the collocation probabilities between each query word and the average of the collocation probabilities between each query word and other words in the matching example sentence except each query word, or, the collocation between each query word The ratio of the sum of the probabilities to the sum of the collocation probabilities between each query word and other words except each query word in the matching example sentence, and the product of the length correction factor, wherein the length correction factor is a length of the matching example sentence related functions.

According to a preferred embodiment of the present invention, the method further includes: if the sentence library is a bilingual sentence library, when displaying each matching example sentence, displaying another language in the bilingual sentence library that is a translation of each matching example sentence example sentences.

According to a preferred embodiment of the present invention, the method further includes: when displaying each matching example sentence, determining and displaying a collocation strength level between each matching example sentence and the query word.

The present invention also provides a device for sorting the retrieval results of example sentences, comprising: a receiving unit for obtaining query words of the user; a retrieval unit for retrieving matching example sentences containing each query word from the sentence database; a calculation unit for In calculating the collocation strength between each matching example sentence and the query word, wherein the collocation strength between the matching example sentence and the query word is divided by the collocation probability between each query word and each query word and the matching example sentence The collocation probability between other words is determined, and the collocation probability between words refers to the possibility of forming a collocation relationship between words; the sorting unit is used to sort each Matching example sentences are sorted.

According to a preferred embodiment of the present invention, the sentence database includes a single sentence database or a double sentence database.

According to a preferred embodiment of the present invention, the collocation strength between the matching example sentences and the query words is equal to: the maximum value of the collocation probabilities between each query word and the probability of each query word and the matching example sentences except each query word The ratio of the maximum value among the collocation probabilities between other words, or the maximum value among the collocation probabilities between each query word and the collocation probability between each query word and other words in the matching example sentence except each query word or the ratio of the average of the collocation probabilities between each query word to the average of the collocation probabilities between each query word and words other than each query word in the matching example sentence, or , the difference between the average value of collocation probabilities between each query word and the average value of collocation probabilities between each query word and other words in the matching example sentence except each query word, or, the difference between each query word The ratio of the sum of collocation probabilities to the sum of collocation probabilities between each query word and other words except each query word in the matching example sentence, and the product of the length correction factor, wherein the length correction factor is one and the matching example sentence Functions related to length.

According to a preferred embodiment of the present invention, the device further includes: a display unit. If the sentence library is a bilingual sentence library, when the display unit displays each matching example sentence, it will display the matching sentences in the bilingual sentence library. An example sentence in another language that is a translation of each other.

According to a preferred embodiment of the present invention, the device further includes: a determining unit, configured to determine a collocation strength level between each matching example sentence and the query when displaying each matching example sentence.

From the above technical solutions, it can be seen that by calculating the collocation strength between each query word in each matching example sentence, and sorting and displaying the matching example sentences according to the collocation strength, the user's language learning purpose and needs can be better met, and the user's language learning can be improved. The user's browsing efficiency is improved, and at the same time, the response times increased by the system to meet the user's needs are reduced.

【Description of drawings】

Fig. 1 is the schematic flow chart of the embodiment of the method for sorting example sentence retrieval results in the present invention;

Fig. 2 is the schematic diagram of embodiment one of example sentence retrieval result display interface in the present invention;

Fig. 3 is the schematic diagram of embodiment two of example sentence retrieval result display interface in the present invention;

Fig. 4 is a schematic structural block diagram of an embodiment of a device for sorting example sentence retrieval results in the present invention.

【detailed description】

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

Please refer to FIG. 1 , which is a schematic flowchart of an embodiment of a method for sorting example sentence retrieval results in the present invention. As shown in Figure 1, the method includes:

Step 101: Obtain the user's query words.

Step 102: Retrieve matching example sentences containing each query word from the sentence database.

Step 103: Calculate the collocation strength between each matching example sentence and the query word.

Step 104: sort each matching example sentence according to the collocation strength between the matching example sentence and the query word.

The above steps are described in detail below.

When a user is learning a language, the purpose of querying one word or multiple words is usually different. When querying a word, the user hopes to obtain an example sentence containing the word in order to understand the usage of the query word in the sentence. When a user queries multiple words, usually these multiple words have a collocation relationship when they are used. The user hopes to obtain example sentences containing these query words, and at the same time wants to know how the collocation relationship between these query words is. reflected in the example sentences. In the embodiment of the present invention, only the collocation relationship of two or more query words in the example sentence will be considered. Therefore, in step 101, the obtained user query words are multiple query words.

In step 102, matching example sentences containing the query word are retrieved from an existing sentence database, which may be a single sentence database or a double sentence database. A single sentence base is a sentence base formed by sentences in one language, and a bilingual sentence base is a sentence base formed by a pair of double sentences, which are composed of sentences in two different languages, and these two sentences are translations of each other . Sentence bases can be generated offline through existing technologies. For example, a single sentence base can be obtained from a large-scale corpus of one language, while a bilingual sentence base can be extracted from a large-scale bilingual corpus. If the sentence database is a bilingual sentence database, when matching example sentences in the source language are retrieved, the corresponding example sentences in the target language can also be obtained accordingly.

In step 103, the collocation strength between the matching example sentence and the query word is determined by the collocation probability between each query word and the collocation probability between each query word and other words in the matching example sentence except each query word. Collocation probability refers to the possibility of forming a collocation relationship between words. For example, "improvement" is often used together with "efficiency", then the probability of collocation between "enhancement" and "efficiency" is relatively high, while "increase" and "area" are rarely used together, then "increase" and " The probability of collocation between "area" is very small. The collocation probability can be obtained through existing technologies, for example, the statistics of the co-occurrence probability between words can be obtained through a large-scale offline corpus, and a language model including the collocation probability between words can be obtained. Since it is a very mature technology to calculate the multivariate co-occurrence probability between words and words in natural language processing, its specific content will not be repeated in the present invention.

The collocation strength between the matching example sentence and the query word is used to measure the closeness of the combination of each query word in the matching example sentence. Using the collocation strength, the application of the same query word in different matching example sentences can be distinguished, so as to find the matching In the example sentences, matching example sentences closely related to each other are returned to the user, and these matching example sentences are usually what the user really wants to obtain.

The collocation strength has multiple implementation modes on the basis of considering the collocation probability between words, one of which is: the collocation strength is equal to the maximum value of the collocation probability between each query word and each query word and the matching example sentence except each The ratio of the maximum value of collocation probabilities between other words other than the query words, or the maximum value of the collocation probabilities between each query word and the ratio between each query word and other words other than each query word in the matching example sentence The difference between the maximum value in the collocation probability. Expressed in the formula as follows:

or

Where Q is the user's query word, E is the matching example sentence, M(Q, E) is the collocation strength between the matching example sentence and the user's query word, w _i or w _j is one of the query words, w _k is the matching example sentence A word other than each query word in , p(w _i , w _j ) or p(w _i , w _k ) is the collocation probability between two words.

The calculation of the collocation strength can also adopt the following method, that is, the collocation strength is equal to the average value of the collocation probability between each query word and the average value of the collocation probability between each query word and other words except each query word in the matching example sentence , or the difference between the average value of collocation probabilities between each query word and the average value of collocation probabilities between each query word and other words in the matching example sentence except each query word. Expressed as a formula:

or

Where Q is the user's query word, E is the matching example sentence, M(Q, E) is the collocation strength between the matching example sentence and the user's query word, w _i or w _j is one of the query words, w _k is the matching example sentence A word other than each query word in , p(w _i , w _j ) or p( _wi , w _k ) is the collocation probability between two words, |Q| is the number of query words of the user, |E| is the number of words contained in the matching example sentence.

In addition, the collocation strength can also be calculated in the following way, that is, the collocation strength is equal to the sum of the collocation probabilities between each query word and the collocation probability between each query word and other words in the matching example sentence except each query word The ratio of the sum and the product of the length correction factor, wherein the length correction factor is a function related to the length of the matching example sentence. Expressed in the formula as follows:

Where Q is the user's query word, E is the matching example sentence, M(Q, E) is the collocation strength between the matching example sentence and the user's query word, w _i or w _j is one of the query words, w _k is the matching example sentence A word other than each query word in , p(w _i , w _j ) or p(w _i , w _k ) is the collocation probability between two words, L(E) is a word related to the length of the matching example sentence The function of is used to prevent the impact of too long or too short sentences on collocation strength.

In addition, the calculation of collocation strength can also be expressed in other ways, which will not be exhaustive here.

In step 104, each matching example sentence is sorted according to the collocation strength between the matching example sentence and the query word calculated in step 103. Each sorted matching example sentence can be further transmitted to other systems or applications.

In another embodiment of the present invention, after step 104, it may further include displaying each sorted matching example sentence. When displaying example sentences, the collocation strength level between each matching example sentence and the query word of the user may be further determined next to each matching example sentence. The specific method is: classify the collocation strength between each query word in each matching example sentence according to the preset threshold range, and further classify the collocation strength level between the matching example sentence and the user's query word when displaying the sorted matching example sentences to show. For example, the matching strength is divided into three levels. When the matching strength is greater than 0.8, the level is strong; when the matching strength is greater than 0.3 and less than 0.8, the level is medium; when the matching strength is less than 0.3, the level is weak. Weak three levels, corresponding matching example sentences are marked with five five-pointed stars, four five-pointed stars and three five-pointed stars.

Please refer to FIG. 2 . FIG. 2 is a schematic diagram of Embodiment 1 of an example sentence retrieval result display interface in the present invention. As shown in Figure 2, to query word: the query (Query) that " improves "+" efficiency " forms, assuming example sentence 1 to example sentence 3 are according to the collocation strength size between each query word in the matching example sentence calculated in step 103 from Three matching example sentences arranged from high to low, on the right side of each example sentence, there are several five-pointed stars, which are used to identify the level of collocation strength between the matching example sentence and the user's query word, among which the number of five-pointed stars is more than the matching example sentence The matching strength with query words is high. Of course, identifying each matching example sentence with collocation intensity is not limited to the form shown in Fig. 2, and any symbol, number, text or illustration that can be used to represent the size or number should be included in the scope of thought of the present invention Inside.

Please refer to FIG. 3 . FIG. 3 is a schematic diagram of Embodiment 2 of an example sentence retrieval result display interface in the present invention. As shown in Figure 3, if in step 102, the sentence library used to search for matching example sentences is a bilingual sentence library, when displaying each matching example sentence after sorting, another translation of each matching example sentence in the bilingual sentence library will also be displayed. Example sentences in one language.

In Figure 3, compared with Figure 2, there are three English example sentences corresponding to each other as translations below Example Sentences 1 to 3. Of course, the translations of these example sentences can also be placed above them or in other positions that can correspond to them superior.

Please refer to FIG. 4 . FIG. 4 is a structural block diagram of an embodiment of an apparatus for sorting example sentence retrieval results in the present invention. As shown in FIG. 4 , the device includes: a receiving unit 201 , a retrieving unit 202 , a calculating unit 203 , a sorting unit 204 , a displaying unit 205 and a determining unit 206 .

Wherein the receiving unit 201 is used to obtain the user's query words.

When a user is learning a language, the purpose of querying one word or multiple words is usually different. When querying a word, the user hopes to obtain an example sentence containing the word in order to understand the usage of the query word in the sentence. When a user queries multiple words, usually these multiple words have a collocation relationship when they are used. The user hopes to obtain example sentences containing these query words, and at the same time wants to know how the collocation relationship between these query words is. reflected in the example sentences. In the embodiment of the present invention, only the collocation relationship of two or more query words in the example sentence will be considered, so the user query words acquired by the receiving unit 201 are multiple query words.

The retrieval unit 202 is configured to retrieve matching example sentences containing each query word from the sentence database.

The sentence base can be a single sentence base or a double sentence base. A single sentence base is a sentence base formed by sentences in one language, and a bilingual sentence base is a sentence base formed by a pair of double sentences, which are composed of sentences in two different languages, and these two sentences are translations of each other . Sentence bases can be generated offline through existing technologies. For example, a single sentence base can be obtained from a large-scale corpus of one language, while a bilingual sentence base can be extracted from a large-scale bilingual corpus. If the sentence database is a bilingual sentence database, when matching example sentences in the source language are retrieved, the corresponding example sentences in the target language can also be obtained accordingly.

Calculation unit 203, configured to calculate the collocation strength between each matching example sentence and the query word.

The collocation strength between the matching example sentence and the query word is determined by the collocation probability between each query word and the collocation probability between each query word and other words in the matching example sentence except each query word. Collocation probability refers to the possibility of forming a collocation relationship between words. For example, "improvement" is often used together with "efficiency", then the probability of collocation between "enhancement" and "efficiency" is relatively high, while "increase" and "area" are rarely used together, then "increase" and " The probability of collocation between "area" is very small. The collocation probability can be obtained through existing technologies, for example, the statistics of the co-occurrence probability between words can be obtained through a large-scale offline corpus, and a language model including the collocation probability between words can be obtained. Since it is a very mature technology to calculate the multivariate co-occurrence probability between words and words in natural language processing, its specific content will not be repeated in the present invention.

The collocation strength between the matching example sentence and the query word is used to measure the closeness of the combination of each query word in the matching example sentence. Using the collocation strength, the application of the same query word in different matching example sentences can be distinguished, so as to find the matching In the example sentences, matching example sentences closely related to each other are returned to the user, and these matching example sentences are usually what the user really wants to obtain.

The collocation strength has multiple implementation modes on the basis of considering the collocation probability between words, one of which is: the collocation strength is equal to the maximum value of the collocation probability between each query word and each query word and the matching example sentence except each The ratio of the maximum value of collocation probabilities between other words other than the query words, or the maximum value of the collocation probabilities between each query word and the ratio between each query word and other words other than each query word in the matching example sentence The difference between the maximum value in the collocation probability. Expressed in the formula as follows:

or

Where Q is the user's query word, E is the matching example sentence, M(Q, E) is the collocation strength between the matching example sentence and the user's query word, w _i or w _j is one of the query words, w _k is the matching example sentence A word other than each query word in , p(w _i , w _j ) or p(w _i , w _k ) is the collocation probability between two words.

The calculation of the collocation strength can also adopt the following method, that is, the collocation strength is equal to the average value of the collocation probability between each query word and the average value of the collocation probability between each query word and other words except each query word in the matching example sentence , or the difference between the average value of collocation probabilities between each query word and the average value of collocation probabilities between each query word and other words in the matching example sentence except each query word. Expressed as a formula:

or

Where Q is the user's query word, E is the matching example sentence, M(Q, E) is the collocation strength between the matching example sentence and the user's query word, w _i or w _j is one of the query words, w _k is the matching example sentence A word other than each query word in , p(w _i , w _j ) or p( _wi , w _k ) is the collocation probability between two words, |Q| is the number of query words of the user, |E| is the number of words contained in the matching example sentence.

In addition, the collocation strength can also be calculated in the following way, that is, the collocation strength is equal to the sum of the collocation probabilities between each query word and the collocation probability between each query word and other words in the matching example sentence except each query word The ratio of the sum and the product of the length correction factor, wherein the length correction factor is a function related to the length of the matching example sentence. Expressed in the formula as follows:

Where Q is the user's query word, E is the matching example sentence, M(Q, E) is the collocation strength between the matching example sentence and the user's query word, w _i or w _j is one of the query words, w _k is the matching example sentence A word other than each query word in , p(w _i , w _j ) or p(w _i , w _k ) is the collocation probability between two words, L(E) is a word related to the length of the matching example sentence The function of is used to prevent the impact of too long or too short sentences on collocation strength.

In addition, the calculation of collocation strength can also be expressed in other ways, which will not be exhaustive here.

The sorting unit 204 is configured to sort each matching example sentence according to the matching strength between the matching example sentence and the query word. Each sorted matching example sentence can be further transmitted to other systems or applications.

The display unit 205 is configured to display each matching example sentence after sorting. If the sentence base for which the retrieval unit 202 retrieves the matching example sentences is a bilingual sentence base, when displaying each sorted matching example sentence, the display unit 205 will also display an example sentence in another language that is a translation of each matching example sentence in the bilingual sentence base. The determination unit 206 is configured to determine the collocation strength level between each matching example sentence and the query word of the user. The specific method is: according to the preset threshold interval, the collocation strength between each query word in each matching example sentence is graded, and the display unit 205 can further compare the matching example sentence and the user's query word when displaying the sorted matching example sentences. The collocation strength level is displayed. For example, the collocation strength is divided into three grades. When the collocation strength is greater than 0.8, the grade is strong; when the collocation strength is greater than 0.3 and less than 0.8, the grade is medium; when the collocation strength is less than 0.3, the grade is weak; , Weak three grades, corresponding matching example sentences are marked with five five-pointed stars, four five-pointed stars and three five-pointed stars.

Please refer to FIG. 2 . FIG. 2 is a schematic diagram of Embodiment 1 of an example sentence retrieval result display interface in the present invention. As shown in Figure 2, for the query (Query) formed by the query word: "improvement" + "efficiency", it is assumed that example sentences 1 to 3 are arranged according to the collocation strength between each query word in the matching example sentence from high to low Three matching example sentences, on the right side of each example sentence, there are several five-pointed stars, which are used to identify the collocation strength level between the matching example sentence and the user's query word, among which the number of five-pointed stars is more, indicating the matching between the example sentence and the query word The collocation strength is higher. Of course, identifying each matching example sentence with collocation intensity is not limited to the form shown in Fig. 2, and any symbol, number, text or illustration that can be used to represent the size or number should be included in the scope of thought of the present invention Inside.

Please refer to FIG. 3 . FIG. 3 is a schematic diagram of Embodiment 2 of an example sentence retrieval result display interface in the present invention. As shown in Figure 3, if the sentence library for searching matching example sentences at the retrieval unit 202 is a bilingual sentence library, when each matching example sentence after sorting is displayed, another language that is a translation of each matching example sentence in the bilingual sentence library will also be displayed example sentences.

In Figure 3, compared with Figure 2, there are three English example sentences corresponding to each other as translations below Example Sentences 1 to 3. Of course, the translations of these example sentences can also be placed above them or in other positions that can correspond to them superior.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the present invention. within the scope of protection.

Claims (10)

1. A method for sorting example sentence retrieval results, characterized in that, the method comprises: A. Obtain the user's query words; B. Retrieve a matching example sentence that includes the query word from the sentence bank; C. Calculate the collocation strength between each matching example sentence and the query word, wherein the collocation strength between the matching example sentence and the query word is determined by the collocation probability between each query word and each query word and the matching example sentence except each query The collocation probability between words other than words is determined, and the collocation probability between words refers to the possibility of forming a collocation relationship between words; D. Sorting each matching example sentence according to the collocation strength between the matching example sentence and the query word. 2. The method according to claim 1, wherein the sentence database comprises a single sentence database or a double sentence database. 3. The method according to claim 1, wherein the collocation strength between the matching example sentences and the query words is equal to: the maximum value in the collocation probability between each query words and the difference between each query word and the matching example sentences The ratio of the largest values among the collocation probabilities between terms other than each query term, or, the difference between the maximum value of the collocation probabilities between each query word and the maximum value of the collocation probabilities between each query word and words other than each query word in the matching example sentence, or, the ratio of the average of the collocation probabilities between each query word to the average of the collocation probabilities between each query word and words other than each query word in the matching example sentence, or, the difference between the average of the collocation probabilities between each query word and the average of the collocation probabilities between each query word and words other than each query word in the matching example sentence, or, The ratio of the sum of the collocation probability between each query word to the sum of the collocation probability between each query word and other words except each query word in the matching example sentence, and the product of the length correction factor, wherein the length correction factor is a function related to the length of matching example sentences. 4. The method according to claim 1, characterized in that, the method further comprises: if the sentence library is a bilingual sentence library, when displaying each matching example sentence, displaying the interaction between each matching example sentence in the bilingual sentence library An example sentence in another language for the translation. 5 . The method according to claim 1 , further comprising: when displaying each matching example sentence, determining and displaying a collocation strength level between each matching example sentence and the query word. 6. A device for sorting example sentence retrieval results, characterized in that said device comprises: The receiving unit is used to obtain the query words of the user; A retrieval unit is used to retrieve matching example sentences comprising each query word from the sentence database; A calculation unit for calculating the collocation strength between each matching example sentence and the query word, wherein the collocation strength between the matching example sentence and the query word is determined by the collocation probability between each query word and each query word and the matching example sentence The collocation probability between other words except each query word is determined, and the collocation probability between words refers to the possibility of forming a collocation relationship between words; The sorting unit is configured to sort each matching example sentence according to the collocation strength between the matching example sentence and the query word. 7. The device according to claim 6, wherein the sentence database comprises a single sentence database or a double sentence database. 8. The device according to claim 6, wherein the collocation strength between matching example sentences and the query words is equal to: the maximum value of the collocation probability between each query word and the difference between each query word and the matching example sentences The ratio of the largest values among the collocation probabilities between terms other than each query term, or, the difference between the maximum value of the collocation probabilities between each query word and the maximum value of the collocation probabilities between each query word and words other than each query word in the matching example sentence, or, the ratio of the average of the collocation probabilities between each query word to the average of the collocation probabilities between each query word and words other than each query word in the matching example sentence, or, the difference between the average of the collocation probabilities of the query words with each other and the average of the collocation probabilities of each query word with other words in the matching example sentence other than the query words, or, The ratio of the sum of the collocation probability between each query word to the sum of the collocation probability between each query word and other words except each query word in the matching example sentence, and the product of the length correction factor, wherein the length correction factor is a function related to the length of matching example sentences. 9. The device according to claim 6, characterized in that, the device further comprises: a display unit, if the sentence database is a bilingual sentence database, then the display unit displays the bilingual sentence when each matching example sentence is displayed. An example sentence in another language that is a translation of each matching example sentence in the sentence library. 10 . The device according to claim 6 , further comprising: a determination unit configured to determine a collocation strength level between each matching example sentence and the query when displaying each matching example sentence. 11 .