JP6630701B2 - Answer candidate generation device, method, and program - Google Patents

Description

  The present invention relates to an answer candidate generating device, a method, and a program, and more particularly, to an answer candidate generating device, a method, and a program for generating an answer candidate used in an interactive system.

  In general, when talking with a dialogue system such as a robot, the user often speaks and responds. However, recently, there is a kind of chat dialogue from the system side, which is developed on a scenario basis. In the case of an utterance such as asking something from the system, it is necessary to recognize the answer to the other party's answer and speak again. When such a scenario is generated, it is often the case that a possible answer to a question of the system is manually predicted and the scenario is generated.

JP 2014-219822 A

Construction of a natural dialogue platform and its application to speech dialogue toys. Kanako Onishi and others. Materials from the Japanese Society for Artificial Intelligence, SIG-SLUD-B502-14, 2015

  However, since the response pattern of the other party is enormous, it is difficult to cover many of the supposed answers in the interactive system, and there is a problem that the cost is enormous to execute.

  The present invention has been made to solve the above problems, and has as its object to provide an answer candidate generation device, a method, and a program that can obtain appropriate and various answer candidates for a question sentence. .

  In order to achieve the above object, an answer candidate generating apparatus according to a first aspect of the present invention includes: a concept vector model generating unit configured to generate a concept vector representing a concept of the word for each of words included in a text set; A language model generation unit that generates a language model based on N-grams of words included in the set, a sentence pattern generation unit that generates a sentence pattern to which a tag for replacing with a word is added from an input question sentence, An expanded word candidate generation unit that obtains each of the words similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, and sets them as extended answer word candidates; The score of the language model, which is obtained when replacing the tag included in the sentence pattern with an extended answer word candidate, and the extended answer word candidate Total score based on the score of the concept vector of have is configured to include, and extension word candidate reranking unit for obtaining the top N expansion answer word candidates.

  Further, in the answer candidate generation device according to the first invention, the extended word candidate reranking unit may calculate the score of the language model in consideration of the word length of the sentence pattern.

The answer candidate generation device according to the second invention is a concept vector model generation unit that generates a concept vector representing a concept of the word for each of the words included in the text set,
A language model generation unit that generates a language model based on the N-gram of a word included in the text set; and a sentence pattern generation unit that generates a sentence pattern to which a tag for replacing with a word has been added from an input question sentence. An extended word candidate generating unit that obtains each word similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, and sets the word as an extended answer word candidate And integration based on a score of the number of hits obtained as a result of executing a search using a query in which the tags included in the sentence pattern are replaced with extended answer word candidates, and a score of the concept vector for the extended answer word candidates And an extended word candidate reranking unit that obtains the top N extended answer word candidates with a score.

  The answer candidate generation method according to a third aspect, wherein the concept vector model generation unit generates, for each of the words included in the text set, a concept vector representing the concept of the word; A step of generating a language model based on the N-gram of a word included in the text set, and a step of generating a sentence pattern with a tag for replacing the word from the input question sentence by the sentence pattern generation unit; The extended word candidate generation unit obtains each of words similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, Obtained when the extended word candidate reranking unit replaces the tag included in the sentence pattern with an extended answer word candidate. Score of the language model, and wherein the total score based on the score of the concept vectors for the extended answer word candidate is executed includes determining a top N expansion answer word candidates, a.

  In the answer candidate generation method according to a third aspect, the extended word candidate reranking unit may determine the score of the language model in consideration of the word length of the sentence pattern.

  The answer candidate generation method according to a fourth aspect, wherein the concept vector model generation unit generates, for each of the words included in the text set, a concept vector representing the concept of the word; A step of generating a language model based on the N-gram of a word included in the text set, and a step of generating a sentence pattern with a tag for replacing the word from the input question sentence by the sentence pattern generation unit; The extended word candidate generation unit obtains each of words similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, And an extended word candidate reranking unit uses a query obtained by replacing the tag included in the sentence pattern with an extended answer word candidate. Determining the top N extended answer word candidates based on the score of the number of hits obtained as a result of executing the search, and the integrated score based on the concept vector score of the extended answer word candidate. Features.

  A program according to a fifth invention is a program for causing a computer to function as each unit of the answer candidate generation device according to the first and second inventions.

  Further, according to the answer candidate generating apparatus, method, and program of the present invention, a sentence pattern to which a tag for replacing a word is added is generated from an input question sentence, and each of the answer words to the question sentence and the question sentence are generated. Each of the words similar to each of the words included in the is obtained based on the score of the concept vector, as an extended answer word candidate, obtained when replacing the tag included in the sentence pattern with the extended answer word candidate, By obtaining the top N extended answer word candidates having an integrated score based on the score of the language model and the score of the concept vector, it is possible to obtain appropriate and diverse answer candidates for the question sentence.

It is a block diagram showing composition of an answer candidate generation device concerning an embodiment of the invention. It is a figure of the comparative example of the search image of the expansion word in the case where clustering is used and the case where it is not used. 6 is a flowchart illustrating a response candidate generation processing routine in the response candidate generation device according to the embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<Overview according to Embodiment of the Present Invention>

  First, an outline of an embodiment of the present invention will be described.

  In order to solve the above-mentioned problems, in the present embodiment, a user's answer content is predicted for a question sentence made by a dialog system using statistical information using a large amount of text, linguistic information of a question and statistical information. By doing so, support for creating a dialog scenario is realized.

  The answer candidate generation device according to the present embodiment expands an answer word by giving a question sentence from the system and a few words (answer seed) to be an answer of the question sentence, with respect to a scenario of the dialogue system. Is a device that outputs various answers to words.

  In the present embodiment, a large amount of text such as a blog article is statistically processed, and a concept vector expressing the meaning of a word is created. An extended word candidate to be output is determined based on the similarity between the concept vectors. In addition, a sentence pattern is generated by language analysis of the question sentence. By applying extended word candidates output based on concept vectors to sentence patterns and determining whether they are valid as a language model, it is possible to output words that are valid answers to questions with higher accuracy There are features.

  Further, by inputting a set of extended answer words output by the above method as an answer seed again, more diverse extended word candidates can be obtained. At this time, there is a feature that by performing clustering for re-entering extended word candidates with similar meanings and expanding words in each cluster, it is possible to obtain more diverse extended word candidates.

<Configuration of answer candidate generating apparatus according to an embodiment of the present invention>

  Next, the configuration of the answer candidate generation device according to the embodiment of the present invention will be described. As shown in FIG. 1, an answer candidate generating device 100 according to an embodiment of the present invention includes a CPU, a RAM, a ROM storing a program and various data for executing an answer candidate generating process routine described below, And a computer including The answer candidate generation device 100 functionally includes an input unit 10, a calculation unit 20, and an output unit 50 as shown in FIG.

  The input unit 10 receives input of a question sentence and each of answer words to the question sentence (hereinafter, also referred to as “(original) answer seed”). In the present embodiment, a case will be described as an example where a question sentence “What is the food you want to eat freshly and want to eat?” Is received and “Bread, fried food, rice, noodles, yakiniku” is received as each of the answer words.

  The operation unit 20 includes a concept vector model generation unit 22, a concept vector model 24, a language model generation unit 26, a language model 28, a sentence pattern generation unit 30, an extended word candidate generation unit 32, an extended answer word candidate It includes a re-ranking unit 34, a clustering unit 36, and a condition determining unit 38.

  The concept vector model generation unit 22 generates a concept vector representing a concept of a word for each word included in the text set, and stores the concept vector as a concept vector model 24. The concept vector is based on word appearance information from a text set. In the present embodiment, a concept vector creation method using a neural network (Word2vec) is used. Also, any one such as one using singular value decomposition may be used. However, since the extended word candidate generation unit 32, which will be described later, performs a process of synthesizing a plurality of word vectors, the concept vector adopted by the concept vector model generation unit 22 can be used as a synthesized vector. The method of synthesis is not specified.

  The language model generation unit 26 generates a language model 28 based on N-grams of words included in the text set. At this time, the method of creating the language model is not particularly specified, but auxiliary verbs such as “ha” and “ga” are replaced with “*” in order to use the model efficiently. However, the language model 28 created here is in a format that can use a sentence pattern generated by a sentence pattern generation unit 30 described later.

  The sentence pattern generation unit 30 generates a sentence pattern to which a tag for replacing with a word has been added from the question sentence received by the input unit 10. Specifically, first, the sentence pattern generation unit 30 parses the input question sentence. The syntax analysis here is dependency analysis. First, a 5W1H expression such as “why”, “what”, “when”, “where”, and “who” is searched for the question text subjected to the dependency analysis. The corresponding expression of 5W1H is replaced with a tag (<cand>) for replacing with a word. Next, a phrase containing a verb or a verb stem is found from the result of the dependency analysis. With this clause as a reference clause, the remaining clauses are divided before and after the reference clause, and the clauses are combined according to the dependency relations. Here, when combining with the clause before the reference clause, “<cand> *” including a tag is added before the reference clause (* is an arbitrary particle), and then the clause is determined by the dependency relation. Are synthesized. To combine with the clause after the reference clause, a replacement tag “<cand>” is added to the end of the last clause of the question sentence (punctuation marks and question marks are deleted), and the composition is repeated from there according to the dependency relation.

  The standard phrase in the question "What food do you want to eat freshly?" Is the part you want to eat. In the composition with the clause before this reference clause, two sentence patterns of “<cand> * I want to eat” and “<cand> * I want to eat freshly made” are generated. Three sentence patterns are generated: <cand>, <cand> is the food you want to eat, and <cand> is the food you want to eat. Therefore, a total of five sentence patterns are generated. The replacement tag <cand> may be freely set individually, or a sentence pattern in which a particle such as "ha" is replaced with * may be created.

  The following is an output example of a sentence pattern. In this example, particles such as "wa" are replaced with *.

Sentence pattern 1: <cand> * Sentence pattern 2 you want to eat: <cand> * Freshly made * Sentence pattern 3 you want to eat: Food * <cand>
Sentence pattern 4: Food you want to eat * <cand>
Sentence pattern 5: Freshly made * Food you want to eat * <cand>

  The expanded word candidate generation unit 32 obtains each of words similar to each of the answer words to the question sentence received by the input unit 10 and each of the words included in the question sentence based on the concept vector model 24, and Answer word candidates. Further, for each cluster obtained by the clustering unit 36 described later, each word similar to each of the words belonging to the cluster and each of the words included in the question sentence is obtained based on the concept vector model 24, and expanded. Answer word candidates.

The extended word candidate generation unit 32 generates a candidate word for an output extended answer word using the concept vector of the word of the concept vector model 24. A set of input answer words (answer seeds) is represented by α. A set Q of the result of morphological analysis of a question sentence is prepared. Here, a composite vector of the concept vectors of the words in the set α _{+ q} when a certain word q in Q is added to α

Then, the score of the concept vector of a certain word w can be calculated as in the following equation (1).

... (1)

here,

Is a composite vector created for each of the words q

And word w concept vector

Is the degree of similarity. As a method of calculating the similarity, any method can be used as long as the larger the numerical value is, the higher the similarity is. The above score is calculated for all the words included in the concept vector model, and the top n words when the scores are arranged in descending order are used as the extended answer word candidates in the next extended answer word candidate reranking section 34. .

Further, the extended word candidate generating unit 32, the clustered by the clustering unit 36, for example, because the set Q _k of the extended answer word candidate for each cluster is obtained, for the word w, for calculating the score for each cluster. Then, the top n words obtained when the scores calculated for all clusters are arranged in descending order are used by the next extended answer word candidate reranking unit 34 as extended answer word candidates. Note that a score obtained by summing the scores calculated for each cluster for the word w may be used. Further, a calculation method may be adopted in which the number of words belonging to the cluster to which the extended answer word candidate belongs is introduced as a coefficient in the calculation of the above equation (1). For example, if the number of words in all clusters is 100 and the number of words in a certain cluster is 5, 5/100 may be used as a coefficient.

  The extended answer word candidate re-ranking unit 34 obtains the score of the language model 28 obtained when each of the top n extended answer word candidates replaces the tag included in the sentence pattern with the extended answer word candidate. An integrated score is calculated based on the score of the concept vector calculated for the extended answer word candidate, and the top N extended answer word candidates with the integrated score are obtained. Similarly, after the clustering by the clustering unit 36, the score of the language model 28 obtained when the tag included in the sentence pattern is replaced with the extended answer word candidate for each of the upper n extended word candidates. Then, an integrated score based on the score of the concept vector calculated for the extended answer word candidate is calculated, and the top N extended answer word candidates with the integrated score are obtained. The integrated score after clustering is obtained based on the score of the language model 28 and the score obtained by multiplying the score of the concept vector by using the number of words belonging to the cluster to which the extended answer word candidate belongs as a coefficient.

  Specifically, the extended answer word candidate reranking unit 34 first replaces the part of the sentence pattern replacement tag <cand> with the extended answer word candidate for each of the extended answer word candidates, thereby Is applied to the sentence pattern. Then, the replaced sentence is input to the language model 28, and the score of the language model 28 is calculated. Here, any method of calculating the score of the language model 28 can be used. In the present embodiment, for example, the score Score (w) of the extended answer word candidate w is calculated using the conditional probability of a word as a score in a general language model, as in the following equation (2).

... (2)

  Here, in the above equation (2), w represents a certain extended answer word candidate, and p represents a sentence pattern. In the above equation (2), Score (w, p) is calculated for each combination of the extended answer word candidate w and all sentence patterns p, and the sum is obtained by multiplying length (p) as a weight. length (p) represents the word length of the sentence pattern p. The longer the word length, the greater the weight, and considering the word length, the greater the effect of long sentence patterns. Since a long sentence pattern is a sentence closer to a question sentence including a modifier or the like, a word that is likely to appear in a long sentence pattern is a document that is more relevant to the question sentence.

  Further, as an alternative to the language model 28, a Web search is executed using the sentence pattern p in which the extended answer word candidate w is replaced as a query (q (w, p)), and the number of search hits is used as a score (3) It is also possible to calculate Sore (w, p) as in equation (1).

... (3)

  Here, H (q (w, p)) indicates the number of search hits when performing a Web search using the query q (w, p). At this time, the score Score (w) of the extended answer word candidate w is calculated according to the right side of the above equation (2) using Sore (w, p) calculated by the above equation (3).

  Then, for each of the extended answer word candidates w, final integration is performed using the score Score (w) of the language model 28 and the score of the concept vector calculated for the extended answer word candidate w by the extended word candidate generation unit 32. Calculate the score. Here, as a calculation method of the integrated score, any method can be used as long as both the score of the language model 28 and the score of the concept vector are considered. The average of the score of the language model 28 and the score of the concept vector may be used, or the rank of the score arranged in descending order by the score of the language model 28 and the order of the score arranged in descending order by the score of the concept vector. May be used.

  When the extended answer word candidates are arranged in descending order of the integrated score, the top N items are output to the output unit 50. In addition, a word set used in the next clustering unit 36 is used. Here, the output number n and the final output number N of the extended answer word candidates of the extended word candidate generation unit 32 may be freely set if n> N. In addition, as for the integrated score calculation method when clustering is used, the same calculation method as described above may be used, or in addition to the above calculation method, for example, the extended answer word candidate A calculation method may be adopted in which the number of words belonging to the cluster to which the cluster belongs is introduced as a coefficient. This is because many of the extended answer word candidates output by the extended answer word candidate reranking unit 34 match the answer seed and the question sentence, but include extended answer word candidates that are incorrect. There is. Such an extended answer word candidate is often a word having a tendency greatly different from other outputs, and thus tends to form an independent cluster. If the processing is performed using the number of words as a coefficient, the score of the extended answer word candidate output from a small cluster (a cluster with a small number of words) tends to be small. It is possible not to be done.

  The following is an output example of the extended answer word candidates of the top 10 integrated scores when N = 10.

Curry, fried chicken, hamburger, curry rice, yakisoba, side dish, rice bowl, pasta, tonkatsu, fried

  The following is an example of output of the extended answer word candidates of the top 10 integrated scores further obtained by clustering the above 10 cases and using the extended answer word candidates obtained for each cluster.

  Croquette, fried rice, omelet, gyoza, tarako spa, fried rice, fried rice, katsudon, fried rice, crab cream croquette

  The clustering unit 36 performs clustering based on the concept vector model 24 on a word set including the top N extended answer word candidates and answer words for the question sentence. As for the clustering method, any method may be used. A method of setting the number of clusters in advance (k-means method) or a method of automatically determining the number of clusters (Affinity Propagation) may be used.

  The reason for performing clustering in the present embodiment will be described below.

  In the present embodiment, an extended answer word set in which the answer seed is extended can be obtained by giving the input question sentence and the answer seed, but the extended word candidate generation unit 32 and the extended answer word candidate By executing the processing of the re-ranking unit 34 a plurality of times, it is possible to obtain a more diverse and large set of extended answer word candidates. In this case, the N words output by the extended answer word candidate reranking unit 34 are used as the answer seed for the next extension. By using the set of output extended answer word candidates as the next answer seed, it is possible to obtain a word different from the original answer seed initially input or the previous output, but the N extended answer words output The set of candidates has more words than the original answer seed. In this state, if the extended word candidate generation unit 32 calculates the extended answer word candidate, the composite vector of the concept vectors of the words created from the answer seeds is one in which the meanings of many words are averaged. However, there is a problem that the variety of available words is lost.

  In order to solve the above problem, the clustering unit 36 clusters N words used as answer seeds based on the similarity of the concept vectors of the words of the concept vector model 24, and divides the N words into a plurality of clusters. Then, the extended word candidate generation unit 32 calculates the above equation (1) for each cluster, and obtains the top n extended answer word candidates for all clusters. Then, the extended answer word candidate reranking section 34 calculates the score Score (w) of the extended answer word candidate w for each of all the extended answer word candidates by the above equation (2), and further calculates the integrated score. By doing so, output words after the second loop are determined. When clustering is performed, the word set is obtained by adding the word set α of the original answer seed to N words to be newly used. The N words output by the extended answer word candidate reranking unit 34 are words obtained by expanding the original answer seed, but are a set of more diverse words than the answer seed because the question sentence and the language model are taken into account. When a cluster is created in this state, the cluster is easily dispersed. In addition, some of the N words include words output due to an analysis error, and if a cluster is created with a set of these inappropriate words, the answer output from the second time onward is an inappropriate word. Could be occupied. Therefore, the word set α of the original answer seed is likely to have a similar meaning and is definitely a correct word, so clustering including the original answer seed word set α improves the unity of clusters, In addition, it has a function of preventing a large number of inappropriate clusters from being generated.

  There is a feature that by expanding words for each cluster, it is possible to acquire expanded words with more variety. FIG. 2 shows a comparison between search images for extended words when clustering is used and when clustering is not used. Each black circle in FIG. 2 indicates an answer seed word represented by a concept vector, and an asterisk indicates a composite vector. A range surrounded by a dotted line indicates a range that can be searched for based on the similarity of the combined vector. In this example, it is assumed that the composite vector is represented by the center of gravity of the concept vector of each word.

  The following is an example of cluster output.

Cluster (A): Curry, Curry rice, Yakisoba Cluster (A): Dish dish cluster (U): Pasta cluster (D): Fried, hamburger, rice bowl, tonkatsu, fried

  The condition determination unit 38 repeats the processing of the extended word candidate generation unit 32, the extended answer word candidate reranking unit 34, and the clustering unit 36 until a predetermined condition is satisfied. The condition may be, for example, the number of times or the number of output words.

<Operation of Answer Candidate Generation Apparatus According to Embodiment of Present Invention>

  Next, the operation of the answer candidate generating apparatus 100 according to the embodiment of the present invention will be described. In the answer candidate generation device 100, the concept vector model 24 is obtained in advance by the concept vector model generation unit 22, and the language model 28 is obtained by the language model generation unit 26 in advance. When the input unit 10 receives each of the question sentence and the answer word corresponding to the question sentence, the answer candidate generating device 100 executes an answer candidate generating process routine shown in FIG.

  First, in step S100, from the question sentence received by the input unit 10, a sentence pattern to which a tag for replacing with a word is added is generated.

  Next, in step S102, each of the answer words to the question sentence received by the input unit 10 and each of the words similar to each of the words included in the question sentence are converted into a concept vector model calculated by the above equation (1). The top n words obtained when the scores are arranged in descending order are obtained based on the scores of the concept vectors of the words in 24, and are set as extended answer word candidates.

  In step S104, the score of the language model 28 calculated by the above equation (2) and the concept vector of the concept vector calculated for the extended answer word candidate obtained when the tag included in the sentence pattern is replaced with the extended answer word candidate The top N (n> N) extended answer word candidates with integrated scores based on the scores are obtained.

  In step S106, clustering is performed based on the concept vector model 24 on a word set including the top N extended answer word candidates obtained in step S104 or the previous step S110 and answer words to the question sentence.

  In step S108, for each cluster, each word similar to each of the words belonging to the cluster and each of the words included in the question sentence is converted into the concept vector of the word in the concept vector model 24 calculated by the above equation (1). And the top n words when the scores are arranged in descending order are set as extended answer word candidates.

  In step S110, the score of the language model 28, which is obtained when the tag included in the sentence pattern obtained in step S108 is replaced with the extended answer word candidate obtained for the cluster, is calculated, and the score is calculated by the above equation (2). The integrated score based on the calculated score of the language model 28 and the score of the concept vector calculated for the extended answer word candidate finds the top N extended answer word candidates. Here, the integrated score is obtained based on the score of the concept vector, multiplied by the score of the language model 28 and the number of words belonging to the cluster to which the extended answer word candidate belongs as a coefficient.

  In step S112, it is determined whether a condition for terminating the repetition is satisfied. If the condition is satisfied, the process proceeds to step S114. If not, the process returns to step S106 to repeat the process.

  In step S114, the extended answer word candidates obtained in steps S104 to S112 are output to the output unit 50, and the process ends.

  As described above, according to the answer candidate generating device according to the embodiment of the present invention, a sentence pattern to which a tag for replacing with a word is added is generated and included in each of the answer words to the question sentence and the question sentence. Each of the words similar to each of the words is obtained based on the score of the concept vector, and is set as an extended answer word candidate. The language model obtained when the tag included in the sentence pattern is replaced with the extended answer word candidate is obtained. By obtaining the top N extended answer word candidates whose integrated score is based on the score and the concept vector score of the extended answer word candidate, various answer candidates for use in the interactive system can be obtained.

  By using the method according to the embodiment of the present invention, the scenario creator of the dialogue system can give a question sentence from the system and a few words (approximately five words) as answers to the question as answer seeds. , Various answer words (extended answer word candidates) can be obtained.

  Also, in an open domain dialogue such as a chat dialogue, since the answers to the questions are various, it is difficult for the scenario creator to recall all the answers, but by using the method of the embodiment of the present invention, Since the scenario creator does not need to consider many answers with confidence, there is an advantage that the cost of scenario creation can be reduced. At the same time, in the dialogue scenario, the number of answers that can be predicted increases, and the possibility of being able to provide a satisfactory response to the answers of more users increases, so that the quality and satisfaction of the dialogue system can be expected to be improved.

  Further, the method according to the embodiment of the present invention extends the prediction of the answer content of the user to the dialog scenario. In a dialogue scenario, it is common that the system asks a question, and after the user answers the question, the system receives the user's answer and speaks another word. Also in the embodiment of the present invention, the third utterance of the system for the extended answer is created by the system creator by himself. Here, the third utterance may be created with reference to the answer word output in the embodiment of the present invention. This is because, in the embodiment of the present invention, a large number of answer words are output from various viewpoints, which include not only prediction of the user's answer but also those that can be used as the third utterance. Further, by combining a conventional automatic utterance generation device with the creation of the third utterance, the dialogue scenario itself can be automatically generated.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made without departing from the spirit of the present invention.

  For example, in the above-described embodiment, an example has been described in which the answer candidate generating device is configured to include the clustering unit and the condition determining unit, but the present invention is not limited to this. For example, the calculation unit of the answer candidate generation device includes a concept vector model generation unit, a concept vector model, a language model generation unit, a language model, a sentence pattern generation unit, an extended word candidate generation unit, and an extended answer word candidate. And a re-ranking unit. The input unit accepts a question sentence and an answer word corresponding to the question sentence as inputs, and the extended answer word candidate re-ranking unit outputs the top N extended answer word candidates. You may.

  Also, for example, when a Web search is performed using a sentence pattern p replacing the extended answer word candidate w as a query (q (w, p)) instead of the language model score, and the number of search hits is used as a score. The arithmetic unit of the answer candidate generating device includes a concept vector model generating unit, a concept vector model, a sentence pattern generating unit, an extended word candidate generating unit, and an extended answer word candidate reranking unit. It may be. In addition, the configuration may include a condition determining unit, and the processing of the extended word candidate generating unit and the extended answer word candidate reranking unit may be repeated.

10 Input unit 20 Operation unit 22 Concept vector model generation unit 24 Concept vector model 26 Language model generation unit 28 Language model 30 Sentence pattern generation unit 32 Extended word candidate generation unit 34 Extended response word candidate reranking unit 36 Clustering unit 38 Condition determination unit 50 output unit 100 answer candidate generation device

Claims (7)

For each of the words included in the text set, a concept vector model generation unit that generates a concept vector representing the concept of the word,
A language model generation unit that generates a language model based on an N-gram of a word included in the text set;
A sentence pattern generation unit that generates a sentence pattern with a tag for replacing with a word from the input question sentence,
An expanded word candidate generation unit that obtains each of words similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, and sets them as extended answer word candidates; ,
When the tag included in the sentence pattern is replaced with an extended answer word candidate, the integrated score based on the score of the language model and the concept vector for the extended answer word candidate is the top N extended scores. An extended word candidate reranking section for finding answer word candidates,
An answer candidate generation device including:   The answer candidate generating apparatus according to claim 1, wherein the extended word candidate reranking unit obtains a score of the language model in consideration of a word length of the sentence pattern. For each of the words included in the text set, a concept vector model generation unit that generates a concept vector representing the concept of the word,
A language model generation unit that generates a language model based on an N-gram of a word included in the text set;
A sentence pattern generation unit that generates a sentence pattern with a tag for replacing with a word from the input question sentence,
An expanded word candidate generation unit that obtains each of words similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, and sets them as extended answer word candidates; ,
The score of the number of hits obtained as a result of executing a search using a query in which the tag included in the sentence pattern is replaced with an extended answer word candidate, and the integrated score based on the concept vector score for the extended answer word candidate are: An extended word candidate reranking unit for obtaining top N extended answer word candidates;
An answer candidate generation device including: A concept vector model generating unit, for each of the words included in the text set, generating a concept vector representing the concept of the word;
A language model generation unit that generates a language model based on N-grams of words included in the text set;
A step in which the sentence pattern generation unit generates, from the input question sentence, a sentence pattern to which a tag for replacing with a word is added;
The extended word candidate generation unit obtains each of words similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, Steps to
The extended word candidate reranking unit is based on a score of the language model and a score of the concept vector for the extended answer word candidate, obtained when the tag included in the sentence pattern is replaced with an extended answer word candidate. Obtaining the top N extended answer word candidates with the integrated score;
Answer candidate generation method including   The method according to claim 4, wherein the extended word candidate reranking unit obtains a score of the language model in consideration of a word length of the sentence pattern. A concept vector model generating unit, for each of the words included in the text set, generating a concept vector representing the concept of the word;
A language model generation unit that generates a language model based on N-grams of words included in the text set;
A step in which the sentence pattern generation unit generates, from the input question sentence, a sentence pattern to which a tag for replacing with a word is added;
The extended word candidate generation unit obtains each of words similar to each of the answer words to the question sentence and each of the words included in the question sentence based on the score of the concept vector, Steps to
An extended word candidate reranking unit, the score of the number of hits obtained as a result of executing a search using a query in which the tag included in the sentence pattern is replaced with an extended answer word candidate, and the concept of the extended answer word candidate Obtaining a top N extended answer word candidate whose integrated score based on the vector score;
Answer candidate generation method including   A program for causing a computer to function as each unit of the answer candidate generation device according to any one of claims 1 to 3.