JP2019194759A - Dialogue system reinforcement device and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an interactive system reinforcing device capable of reinforcing an existing system so as to operate in accordance with characteristics according to a specific purpose. SOLUTION: A dialogue system reinforcing device 50 which operates together with a dialogue system 52 and gives a characteristic to dialogue by the dialogue system 52 receives an input 54 given to the dialogue system 52, and outputs a response to the input on behalf of the dialogue system 52. A response propriety determination unit 64 that determines whether or not to respond, and a response that responds to the determination by the response propriety determination unit 64 being affirmative to generate a response by a response generation method prepared in advance based on the input. A generator 72, an output of the response generator 72, and an output of the dialogue system 52 are connected so as to receive the output of the response generator 72 according to whether or not the determination result of the response availability determination unit 64 is affirmative. And a selector 74 for selectively outputting the output of the dialogue system 52. [Selection diagram] Fig. 1

Description

  The present invention relates to a dialogue system, and more particularly, to a technique for giving a feature to an existing dialogue system by cooperating with the existing dialogue system.

  In recent years, dialogue systems have been actively developed due to the development of technologies such as machine learning and deep learning. Various dialogue systems are already available. For example, smartphones and personal computers are not only available in existing systems, but they are similar to personal computers but do not have a monitor or keyboard. There are also commercially available devices that perform the above.

  Since these interactive systems are not limited in use, they can return a reasonable response to any utterance. However, there is plenty of room for improvement when considering use for specific purposes.

  For example, consider a dialogue system that children want to use. It is desirable for such a system not only to communicate with children in a casual manner, but also to communicate various knowledge to children, such as parents and school teachers, for example. In other words, it is desirable to have a dialogue system specialized for education.

  When developing such a dialogue system for education, even if there is an existing general-purpose dialogue system, the content of the response can be changed drastically by replacing the database, or the dialogue system itself can be re-developed. There is a need to. In order to redevelop the dialogue system itself, high costs are required. In addition, if the content of the response is changed significantly in the existing dialog system, the goodness of the dialog system that has been developed may be thrown away.

  Therefore, there is a need for a technology that enables dialogue that matches a specific scene such as for education using an existing dialogue system.

  Japanese Laid-Open Patent Publication No. 2003-228561 discloses a conventional technique for developing a new dialog system that enables a dialog suitable for a specific scene using an existing dialog system. Patent Document 1 discloses a dialogue system that prepares a plurality of databases respectively corresponding to a plurality of specific scenes and selects a dialogue scene based on an instruction from a human. This dialogue system uses a database corresponding to a selected scene to create a response sentence that constitutes a natural dialogue according to the scene, and synthesizes it as a speech and outputs it.

  Further, Patent Document 2 listed below discloses an apparatus that appropriately selects response candidates output from an interactive system suitable for speech input in an interactive response apparatus including a plurality of interactive systems including those whose internal configuration is not clear. Has been.

JP 2004-219714 A JP 2016-212541 A

  However, the following problems remain to be solved in the prior art. In the case of the technique disclosed in Patent Document 1, there is a problem that it cannot be used even if there is an existing system. Since an existing dialog system cannot be used, it is necessary to develop a dialog system capable of dialog suitable for each scene from scratch. For those who already have a dialogue system or who want to utilize an existing dialogue system, the method of Patent Document 1 cannot be adopted because the development cost of the dialogue system is too high. In addition, if an existing interactive system has already started providing services, introducing a completely new interactive system will not only cause cost problems, but will also affect the operation of the interactive system as seen by the user. There is also a problem that large changes are likely to occur and are difficult for the user to use.

  On the other hand, Patent Document 2 proposes a useful method of utilizing an existing dialogue system. However, in Patent Document 2, there is a problem that the range in which an existing dialogue system can be reused is within the range of utilization originally intended by the existing dialogue system. For example, the dialog system cannot be specialized in a genre that is not intended for the existing dialog system.

  Accordingly, an object of the present invention is to provide an interactive system reinforcement device and a program therefor that can reinforce an existing system so as to operate according to characteristics according to a specific purpose.

  An interactive system reinforcement device according to a first aspect of the present invention is an interactive system reinforcement device that operates together with an interactive system and gives a characteristic to an interactive operation by the interactive system, receives an input given to the interactive system, and replaces the interactive system. A first determination unit that determines whether or not to output a response to the input, and a response generation method prepared in advance based on the input in response to a positive determination by the first determination unit It is connected to receive the first response generation means for generating the response, the output of the first response generation means, and the output of the dialogue system, and whether or not the determination result of the first determination means is affirmative Accordingly, it includes first selection means for selectively outputting the output of the first response generation means and the output of the dialogue system.

  Preferably, the first determination unit includes a response sentence candidate storage unit that stores a plurality of response sentence candidates so that a response sentence candidate related to the word can be searched using the word as a key, and a word included in the input from the input Extracted by the word extraction means, the response sentence candidate extraction means for searching for and extracting the response sentence candidates from the response sentence candidate storage means using the word extracted by the word extraction means as a key, and the response sentence candidate extraction means Means for determining whether or not to output a response to the input on behalf of the dialogue system according to whether or not there is a response sentence candidate.

  More preferably, the first response generation means includes a score calculation means for calculating a feature score indicating a degree representing a feature for each of the response sentence candidates extracted by the response sentence candidate extraction means, and a response sentence Means for selecting a response sentence candidate having the highest feature score calculated by the score calculation means from among the response sentence candidates extracted by the candidate extraction means, and outputting the response sentence candidate to the first selection means.

  More preferably, the first response generation unit includes a response sentence candidate storage unit that stores a plurality of response sentence candidates so that a response sentence candidate related to the word can be searched using the word as a key, and is included in the input from the input. Word extraction means for extracting a word to be read, response sentence candidate extraction means for searching for and extracting response sentence candidates from the response sentence candidate storage means using the word extracted by the word extraction means as a key, and extraction by the response sentence candidate extraction means For each of the response sentence candidates, a score calculation means for calculating a feature score indicating a degree representing the feature and a response sentence candidate extracted by the response sentence candidate extraction means, the score calculation means most Means for selecting a response sentence candidate for which a high feature score has been calculated and outputting it to the first selection means.

  Preferably, the first response generation unit includes a specific dialogue data item storage unit that stores a plurality of specific dialogue data items, and each of the specific dialogue data items includes a first response sentence and the first response sentence. A first response sentence to be output after, and at least a first word included in the first response sentence, and the first response generation means further includes a system response sentence output by the dialog system for the input Is connected to the dialogue system so as to receive the word, and the word extraction means for extracting the word included in the system response sentence, and the specific dialogue data item including the word extracted by the word extraction means as the first word Data item extraction means for extracting from the data item storage means, and first selection means for outputting a first response sentence included in the specific dialog data item extracted by the data item extraction means Means for inputting, second determination means for determining whether or not the specific dialogue data item has been extracted by the data item extraction means, and for storing an output history indicating that the first response sentence has been output The first determination means refers to the history storage means in response to the input being given to the dialogue system, and the input is input immediately after the first response sentence is output. And determining whether or not to output a response to the input in place of the dialog system, the dialog system reinforcement device further includes a dialog system by the second determination means. In response to determining that the input given to the input is the input immediately after the output of the first response sentence, based on the second response sentence in the same specific interaction data item as the first response sentence Generate a response Second response generating means for performing the operation, wherein the first selecting means is connected to receive the output of the first response generating means, the output of the dialogue system, and the output of the second response generating means. The output of the first response generating means, the output of the second response generating means, and the output of the dialogue system according to the combination of the determination result of the second determination means and the determination result of the third determination means Second selection means for selectively outputting is included.

  A computer program according to the second aspect of the present invention (hereinafter simply referred to as “program”) causes a computer to function as each unit of any of the above-described apparatuses.

  The technical scope of the present invention and the equivalent scope thereof will become more apparent from the following description of the embodiments and the drawings.

FIG. 1 is a schematic block diagram of an interactive system reinforcement device according to a first embodiment of the present invention. FIG. 2 is a schematic diagram showing the configuration of the news DB used in the first embodiment. FIG. 3 is a flowchart illustrating a control structure of a program that causes a computer to function as the interactive system reinforcement device according to the first embodiment. FIG. 4 is a block diagram illustrating a determination model learning system for calculating a score indicating the likelihood of news of each response candidate in the first embodiment. FIG. 5 is a diagram showing a schematic structure of a multi-column convolutional neural network (MCNN) constituting the determination model shown in FIG. FIG. 6 is a schematic diagram illustrating a schematic configuration and an operation principle of one column of MCNN illustrated in FIG. FIG. 7 is a schematic diagram illustrating a method of generating a second word vector sequence that is a feature input to the determination model illustrated in FIGS. 5 and 6. FIG. 8 is a schematic diagram illustrating a method for generating a third word vector sequence that is a feature input to the determination model illustrated in FIGS. 5 and 6. FIG. 9 is a schematic diagram illustrating a method of generating a fourth word vector sequence that is a feature input to the determination model illustrated in FIGS. 5 and 6. FIG. 10 is a flowchart of a program that causes a computer to function as a news title scoring unit in the interactive system according to the first embodiment of the present invention. FIG. 11 is a flowchart of a program that causes a computer to function as a response generation unit in the interactive system according to the first embodiment. FIG. 12 is a block diagram showing a schematic configuration of a question answering system which is an interactive system reinforcement device according to the second embodiment of the present invention. FIG. 13 is a schematic diagram showing a schematic configuration of the question answer DB shown in FIG. FIG. 14 is a block diagram showing a configuration of the question processing unit shown in FIG. 15 is a block diagram showing a configuration of the state determination unit shown in FIG. FIG. 16 is a block diagram showing a configuration of the answer processing unit shown in FIG. FIG. 17 is a flowchart showing a control structure of a program that causes a computer to function as the question processing unit shown in FIG. FIG. 18 is a flowchart illustrating a control structure of a program that causes a computer to function as the state determination unit and the answer processing unit illustrated in FIG. FIG. 19 is a diagram illustrating an external appearance of a computer system that implements the first and second embodiments. FIG. 20 is a block diagram showing an internal configuration of the computer shown in FIG.

  In the following description and drawings, the same parts are denoted by the same reference numerals. Therefore, detailed description thereof will not be repeated.

<First Embodiment>
[Overview]
The dialogue system reinforcement apparatus according to the first embodiment cooperates with an existing dialogue system so that the existing dialogue system performs a dialogue specialized for news. Specifically, when the user's utterance is a topic used in news, this dialog system reinforcement device gives information about the news and simultaneously outputs a response sentence related to the news. In order to realize this, first, a topic word is extracted from the inputted utterance sentence. If the word is contained in a database containing previously prepared news information or the like (hereinafter simply referred to as “database”), relevant information is acquired from the database. For example, if the input includes the topic of “terrorism”, the news title “French Finance Minister: Monitoring of Terrorist Funds, Confirming Cooperation with Japan” can be acquired. This sentence is regarded as a user's utterance sentence and is input to an existing dialogue system to obtain a response sentence. Using the acquired response sentence and the sentence obtained from the database, for example, there seems to be a news saying “French Finance Minister: Monitoring of Terrorist Funds, Confirming Cooperation with Japan”. The investigative authorities seem to stop the indiscriminate terrorism that occurred in London and Madrid. " As a result, the existing dialog system can be used as a dialog system specialized in news.

[Constitution]
"overall structure"
With reference to FIG. 1, the dialogue system reinforcement device 50 according to the first embodiment operates in cooperation with an existing dialogue system 52 and causes the dialogue system 52 to function as a dialogue system specialized in news. The dialogue system reinforcement device 50 includes a news DB 66 and a dialogue history DB 68. Although the structure of the news DB 66 will be described later, it is a database prepared in advance so that a news title including the word can be taken out using the word as a search key. The news DB 66 stores a large number of news titles.

  The dialogue system reinforcement device 50 further receives a topic 54 that is a natural language sentence, extracts a topic from the input 54, and a news title having a word that matches the topic extracted by the topic extraction unit 60. Is received from the news DB 66 and based on the news title acquisition result by the news title acquisition unit 62, the input 54 is responded by using the function of the dialog system reinforcement device 50, or the dialog A response availability determination unit 64 for outputting a selection signal 76 indicating whether the system 52 is to respond. In the present embodiment, the selection signal 76 takes a first value when it should respond using the function of the interactive system reinforcement device 50, and takes a second value otherwise. The response availability determination unit 64 outputs the news title 78 selected by the news title acquisition unit 62 when it should respond using the function of the dialog system reinforcement device 50.

  The topic refers to a noun included in the input 54, and refers to a noun suitable for identifying a topic, excluding nouns that frequently appear during conversation. A plurality of topics may be extracted from one input 54. In that case, a news title is acquired from the news DB 66 using each of the topics as a key, and the union of these is used. Further, instead of acquiring the news title, the news text may be acquired and the union of them may be used.

  Further, in response to the selection signal 76, the dialog system reinforcement device 50 selects the news title 78 output by the response availability determination unit 64 when the selection signal 76 is the first value, and inputs the input 54 when the selection signal 76 is the second value. A selection unit 70 to select and input as an input 80 after selection to the dialog system 52, a response 82 output from the dialog system 52 and a news output from the response availability determination unit 64 when the selection signal 76 is the first value. The response generator 72 for generating and outputting a response 84 using the title 78, and the response 84 when the selection signal 76 is the first value, and the output of the dialogue system 52 when the selection signal 76 is the second value. And a selection unit 74 for selecting and outputting the responses 82 as responses 56 of the dialog system.

<News DB 66>
Referring to FIG. 2, news DB 66 includes a plurality of records. Each record includes a key and a news title that can be extracted using the key as a search key. The key of each record consists of a word (topic) in the corresponding news title. One record can have a plurality of keys, and if any key matches a keyword for search, the record can be taken out from the news DB 66.

<Program structure>
A program for causing a computer to function as the interactive system reinforcement device 50 has the following control structure. Referring to FIG. 3, this program extracts a word as a topic from input 54, and searches news DB 66 using the topic extracted in step 90 as a search key, and records having the topic as a key. Are acquired from the news DB 66, and a step 94 for executing the following processing 96 on each of the news titles acquired in step 92 is included. If no record is found in step 92, the process 96 is not executed once, and the control proceeds to the step after step 94 (step 98 described later).

  The process 96 refers to the dialog history DB 68 to determine whether or not the news title to be processed is in the recent history, and the determination of step 130 and branching of the control flow according to the result is performed. When the determination is negative, step 132 calculates a score indicating the news likeness of the news title to be processed, and when the determination at step 130 is affirmative, step 134 sets the news likelihood score of the processing target news title to zero. Including. When step 132 or step 134 ends, the process 96 for the news title to be processed ends. If there is no news title acquired in step 92, the process 96 is not executed once as described above, and therefore a news title with a score cannot be obtained.

  This program further determines whether or not there is a news record with a news likelihood score> 0 as a result of step 94, outputs a selection signal 76 indicating the determination result according to the result, and branches the flow of control 98. including. If the determination result is affirmative, the selection signal 76 has a first value, and if not, the selection signal 76 has a second value. In addition, when the determination in step 98 is affirmative, the program selects a news title with the highest score and outputs it to the selection unit 70 as the news title 78 shown in FIG. 1, and the determination result in step 98 is affirmative. When the response 82 output from the dialogue system 52 and the news title 78 are combined to generate a response 84 and output it to the selection unit 74, and when the determination at step 98 is negative, Selecting and providing the input 54 to the dialog system 52 to obtain its output response 82.

  The program further selects a response generated in step 102 when the determination in step 98 is affirmative, and otherwise selects the response obtained in step 104 and outputs it as a response 56. , A response 56, and a step 108 of registering the news title 78 in the dialog history DB 68 as a history. When the determination by the response availability determination unit 64 is negative, there is no news title 78. Therefore, in this case, the input history 54, the response 56, and an empty character string or a symbol indicating that no news title has been selected are registered in the dialog history DB 68 as a history.

《Model learning for news-likeness calculation》
The calculation of the score in step 132 of FIG. 3 is performed using the news likelihood determination model 168 shown in FIG. In the present embodiment, a convolutional neural network which is a kind of so-called deep neural network, more specifically, a multicolumn convolutional neural network (MCNN) is used as the news likelihood determination model 168.

<Model learning unit 162>
FIG. 4 shows the configuration of the model learning device 150 for learning the news likelihood determination model 168. FIG. 5 shows a schematic configuration of the news likelihood determination model 168. FIG. 6 shows the configuration of one column of the news likelihood determination model 168. 7, 8, and 9 are diagrams for explaining a method of deriving a feature that becomes an input to each column of the news likelihood determination model 168.

  Referring to FIG. 4, model learning apparatus 150 refers to news term dictionary 164 and connected word dictionary 166, and referring to these, any of connected words stored in connected word dictionary 166 from large-scale text database 160, A text in which any of the news terms stored in the news term dictionary 164 continuously appears is extracted and a news term included in the text is attached to each text as a label. The text extraction / classification unit 190 that extracts text without any continuous text without labeling, and the labeled text and the unlabeled text output from the text extraction / classification unit 190 are like news. A learning text storage unit 192 that stores the determination model 168 for learning, and a learning text storage unit 19 Using the stored learning text to include a learning processing unit 194 to perform the learning of the news likeness decision model 168. If the text is labeled, the text is a news title; if the text is not labeled, the text is not a news title. In the learning texts stored in the learning text storage unit 192, these are respectively labeled as positive examples and negative examples of texts including news titles. For example, this label has a positive example of (1, 0) and a negative example of (0, 1). Learning of the news likeness determination model 168 by the learning processing unit 194 is performed by the error back propagation method as in a normal neural network except that the model is a multi-column convolutional neural network and what is used as a feature. Done.

  Referring to FIG. 4, step 132 shown in FIG. 3 uses a news title storage unit 170 for storing the news title extracted in step 92 of FIG. The news title scoring unit 172 for calculating the news-likeness score and outputting the scored news title 174 is performed.

  In the present embodiment, the text extraction / classification unit 190 is realized by a program. This program searches the large text database 160 for text that includes a word string in which any of the news terms stored in the news term dictionary 164 appears after any of the connected words stored in the connection word dictionary 166. Generate a search expression, extract all the text that satisfies the condition from the large-scale text database 160 using this search expression, and for each extracted text, extract the text before the connected word And storing the text with the searched news term as a label in the learning text storage unit 192 as learning data.

  Referring to FIG. 5, MCNN 220 constituting news likelihood determination model 168 includes a neural network layer 230 including first to fourth convolutional neural network groups 260, 262, 264, and 266, and each of the neural network layers 230. A connection layer 232 that linearly connects the outputs of the neural network, and a softmax function applied to a vector output from the connection layer 232, a score between 0 and 1 indicating whether the news title candidate is a true news title. And a Softmax layer 234 that is evaluated and output at

  The first convolutional neural network group 260 includes a first column sub-network that receives a first word vector sequence to be described later. The second convolutional neural network group 262 includes second, third, and fourth column sub-networks that respectively receive three second word vector sequences to be described later. The third convolutional neural network group 264 includes fifth, sixth, seventh, and eighth column sub-networks that respectively receive four third word vector sequences described later. The fourth convolutional neural network group 266 includes ninth, tenth, and eleventh column sub-networks that receive three fourth word vector sequences to be described later. These sub-networks are all convolutional neural networks.

  The output of each convolutional neural network in the neural network layer 230 is simply linearly connected in the connection layer 232 and becomes an input vector to the Softmax layer 234.

  The function of the MCNN 220 will be described in detail. FIG. 6 shows the configuration of the sub-network 280 that constitutes one column of the third convolutional neural network group 264 as a representative. The other columns in the third convolutional neural network group 264 and the columns included in the other convolutional neural network groups 260, 262, and 266 have the same structure as that shown in FIG. Here, in order to make the explanation easy to understand, it is assumed that the sub-network 280 includes only the input layer 290, the convolution layer 292, and the pooling layer 294. However, the sub-network 280 may include a plurality of these three layers. Good.

In the input layer 290, word vector strings X ₁ , X ₂ ,..., X _{| t |} obtained by converting each word of the feature obtained from the news title into a word vector in the learning data are input. . The word vector sequence _{X 1, X 2, ...,} X | t | , the matrix _{T = [X 1, X 2} , ..., X | t |] expressed as ^T. M feature maps are applied to this matrix T. The feature map is a vector, and the vector O, which is an element of each feature map, applies an N-gram 310 while applying a filter indicated by f _j (1 ≦ j ≦ M) to an N-gram consisting of consecutive word vectors. Calculated by moving. N is an arbitrary natural number, but N = 3 in this embodiment. That is, O is represented by the following formula.

なお、素性マップの全体にわたりＮを等しくしてもよいし、異なるものがあってもよい。Ｎとしては、２、３、４及び５程度が適当である。本実施の形態では、重み行列は各カラムセット内の畳み込みニューラルネットワークにおいて等しくしてある。これらは互いに異なっていても良いが、実際、このように等しくした方が、各重み行列を独立に学習する場合より精度が高くなる。なお、畳み込みニューラルネットワークで使用するフィルタとしては、どのようなものを用いることもできる。画像処理で用いるフィルタを流用してもよい。

Note that N may be equal throughout the entire feature map, or there may be different ones. As N, 2, 3, 4 and 5 are appropriate. In the present embodiment, the weight matrix is equal in the convolutional neural network in each column set. These may be different from each other, but in fact, the equality is more accurate than the case where each weight matrix is learned independently. Any filter can be used as a filter used in the convolutional neural network. A filter used in image processing may be used.

For each feature map, the next pooling layer 294 performs so-called max pooling. That is, the pooling layer 294 selects, for example, the largest element 320 among the elements of the feature map f _M and takes it out as the element 330. By performing this for each of the feature maps, the elements 330,..., 332 are taken out and connected in the order of f ₁ to f _M to be output as a vector 342 to the connection layer 232. Each of the convolutional neural networks outputs the vectors 340,..., 342,. The connection layer 232 simply connects the vectors 340,..., 342,. As the pooling layer 294, it is said that the one that performs max pooling is more accurate than the one that employs an average value. However, of course, an average value may be adopted, and other representative values may be used as long as the properties of lower layers are well expressed.

<Word vector sequence for learning>
The learning data includes the above 11 types of word vector strings and labels indicating whether the text from which the word vector strings are obtained is a positive example or a negative example.

  The word strings to be input to MCNN are as follows. In this table, TNP represents a clause including a noun included in a news title candidate, and PRED represents a clause including a predicate of a predicate template included in the news title candidate. A predicate template refers to a language unit that is combined with a noun and constitutes a phrase. Specific examples will be described later. In the case of a negative example, a word vector string is generated using, for example, the subject of a sentence instead of the word corresponding to the news word.

これら単語列はいずれも、文全体から抽出される。なお、本実施の形態では、ニュースタイトル候補のテキストは所定の複数文という場合もある。したがってＴＮＰとＰＲＥＤが別文に存在することもあり得る。本実施の形態では、そうした場合でも上記した単語列が得られるようにするために、これら複数文の全ての係り受け木のルートを仮想的なルートノードで互いに結合し、１つの係り受け木を用いて上記単語列を得るようにした。

All of these word strings are extracted from the entire sentence. In this embodiment, the news title candidate text may be a predetermined plurality of sentences. Therefore, TNP and PRED may exist in different sentences. In this embodiment, in order to obtain the above-described word string even in such a case, the roots of all the dependency trees of the plurality of sentences are connected to each other by a virtual root node, and one dependency tree is obtained. Used to obtain the above word string.

  Referring to FIG. 7, the first word string includes a pair of a clause 370 including a noun of a news term extracted from a news title candidate 360 and a predicate 372 of a predicate template. In the present embodiment, word embedding vectors are used as all of the following word vectors in order to preserve the appearance order of words and reduce the amount of calculation.

  Referring to FIG. 7, the word vector sequence of column numbers 2-4, which is the second type of word vector sequence, is based on the order of appearance of the word sequence in the news title candidate 360 from the beginning of the sentence to the news title candidate noun. Includes a word string 380 immediately before the clause 370 including, a word string 382 between the clause 370 and the predicate 372, and a word string 384 from the word after the predicate 372 to the end of the sentence. Therefore, the second word vector sequence is obtained as three word embedded vector sequences.

  Referring to FIG. 8, the word vector sequence in column 5-8, which is the third word vector sequence, is based on the dependency tree of news title candidate 360, and subtree 420 related to clause 370 including the noun of the news title candidate. , A word tree obtained from the subtree 410 of the destination of the predicate 372, the dependency path 422 between the clause 370 and the predicate 372, and the other 424. Therefore, in this example, the third word vector sequence is obtained as four word embedded vector sequences.

  Referring to FIG. 9, the word vector sequence in column 9-11, which is the fourth word vector sequence, is the word sequence in the phrase including predicate 372 and word sequence 440 before predicate 372 in news title candidate 360. And a subsequent word string 442. Therefore, in this case, the fourth word vector sequence is obtained as three word embedded vector sequences.

  At the time of learning, each column of MCNN 220 is given an input vector to each column, the output of MCNN 220 is compared with the label of the text, and the value of the error function is decreased in the normal error back propagation method. Each weight and bias value constituting the MCNN 220 is corrected.

  In the present embodiment, the MCNN 220 has two outputs. Since the output is implemented by the SoftMax function, the first output represents the probability that the news title candidate is a news title, and the second output represents the probability that the news title candidate is not a news title. In the present embodiment, the probability of being a news title is used as the score of each news title.

  Step 132 shown in FIG. 3, that is, the news title scoring unit 172 shown in FIG. 4, is realized by a program that cooperates with the computer. Referring to FIG. 10, the program includes a step 460 for executing the following processing 462 for each news title stored in news title storage unit 170.

  The process 462 includes a step 470 of generating input parameters (first to fourth word vector sequences) to the news likelihood determination model 168, a step 472 of inputting these parameters to the news likelihood determination model 168, and a news likelihood determination model 168. Step 474 of attaching a label (probability of being a news title) to the news title as a score.

<< Response generator 72 >>
The response generation unit 72 shown in FIG. 1 is also realized by a program that causes the computer to function as the response generation unit 72 in cooperation with the computer. Referring to FIG. 11, the program includes a step 500 of randomly selecting one of response forms for responses prepared in advance. This response form is, for example, “There is news [First slot]. [Second slot]”. The news title is assigned to the first slot, and the output of the dialogue system 52 is assigned to the second slot.

  The program further includes a step 502 for substituting the news title in the first slot of the response form, and a step 504 for substituting the response output by the dialog system 52 in the second slot of the response form.

[Operation]
The dialogue system reinforcement device 50 according to the first embodiment operates as follows. The dialog system reinforcement device 50 differs depending on whether or not a news title that matches the topic of the input 54 is stored in the news DB 66 (when it conforms) and when it does not (when it does not conform). Hereinafter, the operation of the interactive system reinforcement device 50 when first adapted and then non-conforming will be described. Prior to this operation, it is assumed that learning of the news likelihood determination model 168 shown in FIG. 4 has been completed.

<< Operation of Dialog System Reinforcing Device 50 When Adapted >>
Referring to FIG. 1, the dialog system reinforcing device 50 operates as follows at the time of adaptation. When the input 54 is input, the topic extraction unit 60 extracts the topic from the input 54 and gives it to the news title acquisition unit 62 (step 90 in FIG. 3 (the same applies hereinafter)). The news title acquisition unit 62 acquires a news title having a word that matches the topic extracted by the topic extraction unit 60 from the news DB 66 and gives it to the response availability determination unit 64 (step 92). The response availability determination unit 64 indicates whether to respond to the input 54 using the function of the dialog system reinforcement device 50 or to respond to the dialog system 52 based on the news title acquisition result by the news title acquisition unit 62. A selection signal 76 is output.

  At this time, the response availability determination unit 64 operates as follows. Referring to FIG. 3, response availability determination unit 64 determines whether or not a history including the news title exists in dialog history DB 68 for each news title given from news title acquisition unit 62 (step 130). ). If there is any (YES in step 130), the news-likeness score of the news title is set to 0 (step 134). If not (NO in step 130), the news-likeness score of the news title is calculated using the news-likeness determination model 168 and the news title scoring unit 172 shown in FIG. 4, and attached to the news title and output.

  Referring to FIG. 3 again, if there is a news title with a score> 0 (YES in step 98), the news title with the highest score is input to the dialog system 52 (step 100). At the time of adaptation, the determination in step 98 is YES, and step 100 is executed. At this time, the selection signal 76 shown in FIG. 1 becomes the first value. The selection unit 70 selects the news title 78 instead of the input 54 and gives it to the dialogue system 52 as the input 80 after selection. The dialogue system 52 outputs a response 82 in response to the selected input 80.

  In step 102 of FIG. 3, the response generation unit 72 generates a response 84 by assigning the news title 78 to the first slot and the response 82 to the second slot of the randomly selected response form, and gives the response 84 to the selection unit 74. . Since the value of the selection signal 76 is the first value, the selection unit 74 selects the response 84 instead of the response 82 and outputs it as the response 56.

  That is, at the time of adaptation, the response of the dialogue system 52 to the input 54 is not output. Rather, a news title that matches the topic contained in the input 54 is selected from the news DB 66 and that news title 78 is provided to the dialog system 52 instead of the input 54. The response of the dialog system 52 at this time is given to the response generation unit 72, and the response 84 is generated and output in combination with the news title 78.

  For example, assume that the input 54 is “I'm afraid of terrorism” and the news title is “French finance minister: monitoring terrorist funds, confirming cooperation with Japan”. In addition, a randomly selected response form is “[First slot] seems to be news. [Second slot]”, and the response 82 from the dialogue system 52 at this time is “monitoring terrorist organizations”. The investigation authorities seem to stop the indiscriminate terrorism that happened in London and Madrid. " Then, the final response is “News from the French Finance Minister: Monitoring of Terrorist Funds, Confirming Cooperation with Japan”. It seems to stop such indiscriminate terrorism. "

<< Operation of Dialog System Reinforcing Device 50 at Non-conformance >>
At the time of non-conformity, the dialogue system reinforcement device 50 operates as follows. The operation from the input 54 to the operation of the response availability determination unit 64 is the same as that at the time of adaptation. However, at the time of non-conformity, the determination at step 98 in FIG. 3 is NO, and the selection signal 76 shown in FIG. 1 has the second value. As a result, the selection unit 70 selects the input 54 instead of the news title 78 and inputs it to the dialogue system 52 as the input 80 after selection. Since the selection signal 76 is the second value, the response generator 72 does not operate. Similarly, since the selection signal 76 is the second value, the selection unit 74 selects the response 82 instead of the response 84 and outputs it as the response 56.

  In other words, in this case, if a related news title is not found, the response is “strengthening surveillance against terrorist organizations and investigating authorities to stop indiscriminate terrorism that occurred in London and Madrid”. Will be output.

  As described above, according to the dialog system reinforcing device 50 according to this embodiment, by adding the dialog system reinforcing device 50 to the existing dialog system, the dialog characterized by the response in a specific field such as news. A system can be realized. Although an embodiment of the news has been described here, for example, sports-related news, entertainment-related news, science and technology field news, political news, etc. Applicable form. Furthermore, as long as electronic data summarized as a news title regarding a specific region / era is available, not only news, an existing dialog system can be operated as a dialog system specialized for such information. At this time, it is not necessary to modify the existing dialog system or make a new one.

<Second Embodiment>
[Overview]
The dialogue system reinforcement apparatus according to the second embodiment of the present invention cooperates with an existing dialogue system so that the existing dialogue system performs a dialogue specialized for education. Specifically, when an educationally important word appears in the middle of a dialogue, a question related to the word is presented and the user is made to answer. In order to be able to ask questions, a question answer database is created in advance using textbook data. A topic is extracted from the user's utterance. When a noun included in a topic is included as a key in the question answer database, a question sentence related to the word is extracted from the question answer database. The user's utterance sentence itself is input to an existing dialogue system to obtain a response sentence. A response sentence is generated by combining the question sentence extracted from the question answer database and the response sentence from the dialogue system, and output. In response to user input after the question text is output, a word that is a topic is extracted, and whether or not the word is the answer to the question (question text) that was presented immediately before is checked. A sentence is generated according to the correct answer. Also, the corresponding answer sentence is extracted from the question answer database. A response sentence is generated and output by combining a sentence according to whether the answer is correct or incorrect and an answer sentence extracted from the question answer database.

[Constitution]
"overall structure"
Referring to FIG. 12, dialogue system reinforcement apparatus 530 according to the second embodiment is connected to receive input 54 to existing dialogue system 52 and the output (response) of dialogue system 52 to this input 54. When an educationally important word appears during the dialogue by the dialogue system 52, a question related to the word is given to the user, and the user is made to answer. Further, the dialog system reinforcement device 530 determines whether the input 54 input from the user immediately after this question is a correct answer to the immediately preceding question, and a sentence according to whether the answer is correct or incorrect and an answer to the question A response sentence is generated by combining sentences and output. That is, unlike the interactive system reinforcement device 50 according to the first embodiment, the interactive system reinforcement device 530 according to the second embodiment is capable of providing some kind of educational to the user through two dialogues with the user. The function of the existing dialogue system 52 is reinforced so as to convey knowledge.

  The dialogue system reinforcement device 530 includes a state determination unit 542 that determines whether the input 54 is an answer to the question in response to the input 54 being given, and outputs a determination signal 560. The determination signal 560 is a first value when the input 54 is an answer to the question, and is a second value otherwise. The dialogue system reinforcement device 530 further includes a dialogue history DB 544 that stores a history of dialogues by the dialogue system 52, a question answer DB 548 that stores a large number of pairs of questions and their answers created from textbook data, etc., and a determination signal In response to 560, a branching unit 540 for branching is included, which includes branching input 54 to input 564 when determination signal 560 is the first value, and branching to input 562 otherwise.

  The dialog system reinforcement device 530 further receives a response 566 that the dialog system 52 outputs to the input 562 when the value of the determination signal 560 is the second value, and uses the dialog history DB 544 and the question answer DB 548, It is determined whether or not a question related to the input 54 can be presented to the user. When possible, a question 570 is generated and a determination signal 572 (first value) is output. In response to receiving the input 564 from the branching unit 540 when the determination signal 560 is the first value and the question processing unit 550 for outputting (second value), the dialogue history DB 544 and the question answer DB 548 An answer processing unit 574 for generating and outputting an answer 576 to the question output by the dialog system reinforcement device 530 immediately before, and the dialog system 52 The response 566, the response (question) 570 from the question processing unit 550, and the response 576 from the answer processing unit 574 are selected, and one of these is selected according to the combination of the values of the determination signal 560 and the determination signal 572 And a selection unit 554 that outputs the response 532.

  Referring to FIG. 13, question answer DB 548 includes a plurality of records. Each record includes a question sentence, an answer to the question sentence, a topic of the question sentence, and an answer sentence to the question sentence. In the present embodiment, each record includes a question sentence and an answer sentence regarding art / history. The answer is simply a noun indicating the answer to the question sentence. A topic refers to a noun included in a question sentence. Also in this embodiment, words that are frequently used are not used as topics, and only words that are suitable for use as topics are used as topics. The answer sentence has a format suitable for being presented to the user as an answer sentence to the question sentence.

  For example, the first record is "Do you know who drew Mona Lisa?" And the answer is "Da Vinci drew Mona Lisa." The answer to this record is “Da Vinci” and the topic is “Mona Lisa”. The configuration of the second and subsequent records is the same. Each record in the question answer DB 548 can be easily generated by extracting a sentence with an appropriate pattern from text data of a textbook, for example. Of course, these records may be prepared manually.

  In the dialogue history DB 544, for example, a plurality of recent questions and user inputs are accumulated. This dialogue history DB 544 is used for determination as to whether or not the immediately preceding response 532 is a question sentence. The dialogue history DB 544 is also used to make a determination as to whether the same question has been made in the past. When the same question as the question made in the past is generated by the dialogue system reinforcement device 530, the question is not output. This is to prevent the dialogue system from moving around repeatedly by repeating the same question. However, since this function is not directly related to the present invention, its details are not described here. However, this function should be easily realized by those skilled in the art.

<< Question Processing Unit 550 >>
Referring to FIG. 14, question processing unit 550 includes topic extraction unit 600 for extracting a word as a topic from response 566, and a question answer record including a question answer record including the word extracted by question processing unit 550 as a topic. Based on the question answer acquisition unit 602 to be acquired from the DB 548 and the question answer record extracted by the question answer acquisition unit 602, whether or not the dialog system reinforcement device 530 should generate and output a question is output as a response of the dialog system 52. Based on the record of the question answer acquired by the question answer acquiring unit 602 when it is determined that the question should be generated by the response enable / disable determining unit 604 for determining whether to output, and the response enable / disable determining unit 604 And a response generation unit 606 for generating a response (question) 570 by the dialog system reinforcement device 530.

<< State determination unit 542 >>
Referring to FIG. 15, state determination unit 542 extracts a topic extraction unit 630 for extracting a word representing a topic from input 54, and a question answer record having the word extracted by topic extraction unit 630 as a topic. Judgment indicating whether or not to generate a response (question) by the dialogue system reinforcement device 530 based on the dialogue history acquired from the answer DB 548 and stored in the dialogue history DB 544 and the acquired record of the question answer And an answer determination unit 632 that outputs a signal 560. The answer determination unit 632 extracts a topic from the topic and the input 54 based on the conversation history DB 544 when the output of the previous dialog system reinforcement device 530 is not a question or when the output of the previous dialog system reinforcement device 530 is a question. When the topic extracted by the unit 630 does not match, it is determined that the input 54 is not an answer to the question, and the determination signal 560 is set to the second value. The output of the previous dialog system reinforcement device 530 is a question, and when the topic matches the topic extracted by the topic extraction unit 630 from the input 54, the determination signal 560 is set to the first value. That is, the determination signal 560 has a first value when it is determined that the question should be generated by the interactive system reinforcement device 530, and has a second value otherwise.

<Answer processing unit 574>
Referring to FIG. 16, in response to the input 564 being given, the answer processing unit 574 refers to the dialog history DB 544 and the question answer DB 548 to pair with the question generated by the dialog system reinforcement device 530 immediately before. The answer sentence extraction unit 660 for extracting the answer sentence that has been created from the question answer DB 548, the answer sentence extracted by the answer sentence extraction unit 660, and the previous question stored in the dialogue history DB 544 are compared with each other. And a response generation unit 662 that generates an answer 576 in an appropriate form according to the determination result.

<< Selection Unit 554 >>
The selection unit 554 selects the response 566 when the value of the determination signal 560 is the second value and the value of the determination signal 572 is the second value. The selection unit 554 selects the response (question) 570 when the value of the determination signal 560 is the second value and the value of the determination signal 572 is the first value. When the value of determination signal 560 is the first value, selection unit 554 selects answer 576 regardless of the value of determination signal 572.

<< Realization of Question Processing Unit 550 by Computer >>
The question processing unit 550 illustrated in FIG. 14 is realized by a program that causes a computer to function as the question processing unit 550 in cooperation with the computer. Referring to FIG. 17, the program obtains in step 690 that obtains response 566 (see FIG. 12) output from dialogue system 52, step 692 that extracts a word representing a topic from response 566, and step 692. Step 694 that searches the question answer DB 548 by using the retrieved topic as a search key and extracts the corresponding record, and step 696 that branches whether the control flow is branched by determining whether or not there is a record extracted in step 694.

  This program is further executed when the determination in step 696 is affirmative, and calculates the score of the question sentence of each record extracted in step 694, and the question sentence in which no conversation history is stored in the conversation history DB 544 And a step 700 for selecting the question sentence having the highest score calculated in step 698, a response 566 from the dialog system 52, and the question sentence selected in step 700, and generating and outputting a response sentence. Step 702 and step 704 for selecting and outputting the response 566 of the interactive system 52 when the determination at Step 696 is negative. The score of the question sentence in step 698 can be calculated by the same method as the calculation of the news-likeness score in the first embodiment. Therefore, details thereof will not be repeated here.

<< Realization of State Determination Unit 542 and Answer Processing Unit 574 by Computer >>
The state determination unit 542 and the answer processing unit 574 illustrated in FIGS. 15 and 16 are realized by a program that causes the computer to function as the state determination unit 542 and the answer processing unit 574 in cooperation with the computer.

  Referring to FIG. 18, this program obtains a dialogue history from dialogue history DB 544 and, based on the dialogue history obtained in step 740, determines whether or not the immediately preceding response is a question sentence, and the control flow Branch 742, step 758 for inputting the input 54 to the dialog system 52 when the determination in step 742 is negative, and step 760 for obtaining a response obtained from the dialog system 52 as a result of the process of step 758; including.

  The program further calculates a score indicating the likelihood of answer to the question of the input 564 from the branching unit 540 when the determination in step 742 is affirmative, and the score calculated in step 744 is based on the threshold value. And step 746 for branching the control flow according to the determination. The score calculation in step 744 can be calculated by the same method as the news-likeness score in the first embodiment. If the determination in step 746 is negative, control proceeds to step 758 described above.

  The program further extracts a word as a topic from the input 564 when the determination at step 746 is affirmative, and reads a record having the topic extracted at step 748 from the question answer DB 548 and retrieves the answer sentence 750. including.

  The program further determines whether the topic extracted in step 750 matches the topic of the question output by the dialog system reinforcement device 530 immediately before, and branches control according to the determination result, step 752; When the determination in 752 is affirmative, step 754 for generating a response that the input 54 is a correct answer to the question using the answer sentence extracted in step 750, and input when the determination in step 752 is negative 54 is a wrong answer to the question, and a response that the correct answer is the answer sentence extracted in step 750 is generated. Step 756, and the response generated in these steps after step 760, step 754, and step 756 And step 762 of outputting as a response (question) 570 shown in FIG.

[Operation]
The dialogue system reinforcement device 530 according to the second embodiment operates roughly in any of the following three ways. The first is a case of responding to the input 54 without particularly asking a question (normal time). The second is a case where a question is asked to the input 54 (at the time of a question). The third is a case where a response is made as an answer to the input 54 immediately after the question is made (when answering). Hereinafter, these three cases will be described.

"Normal time"
Normally, the input 54 is given to the state determination unit 542 and the branching unit 540. At this time, the value of the determination signal 560 output from the state determination unit 542 becomes the second value. That is, with reference to FIG. 15, the topic extraction unit 630 extracts topics from the input 54. The answer determination unit 632 reads a question answer record that matches this topic from the question answer DB 548. In normal times, the response output immediately before by the dialog system reinforcement device 530 is not a question. Therefore, the answer determination unit 632 sets the determination signal 560 to the second value.

  Accordingly, branching unit 540 provides input 54 to dialog system 52 as input 562. The dialog system 52 outputs some response 566 in response to this input 562. This response 566 is given to the question processing unit 550 and the selection unit 554.

  Referring to FIG. 14, topic extraction unit 600 extracts a word representing a topic from response 566. The question answer acquisition unit 602 acquires from the question answer DB 548 a question answer record that includes the word extracted by the topic extraction unit 600 in the topic. The response availability determination unit 604 determines whether the dialog system reinforcement device 530 should generate and output a question or output the response of the dialog system 52 based on the question answer record extracted by the question answer acquisition unit 602. To do. Since this example is a normal time, the response availability determination unit 604 determines that the response of the interactive system 52 should be output. As a result, the value of the determination signal 572 output from the response availability determination unit 604 becomes the second value.

  Since the value of the determination signal 560 is the second value and the value of the determination signal 572 is the second value, the selection unit 554 selects the response 566 and outputs it as the response 532.

<At the time of question>
Even at the time of a question, the operation of the state determination unit 542 operates in the same manner as in the normal time.

  Input 54 is thus provided as input 562 to dialog system 52, which outputs response 566. The response 566 is given to the question processing unit 550.

  Referring to FIG. 14, topic extraction unit 600 extracts a word representing a topic from response 566. The question answer acquisition unit 602 acquires from the question answer DB 548 a question answer record that includes the word extracted by the topic extraction unit 600 in the topic. Here, it is understood from the dialogue history stored in the dialogue history DB 544 that the output of the previous dialogue system reinforcement device 530 is not a question. It is assumed that a question answer record is extracted from the question answer DB 548. Then, the response availability determination unit 604 determines that the dialogue system reinforcement device 530 should generate and output a question based on the question answer record extracted by the question answer acquisition unit 602. Therefore, the determination signal 572 becomes the first value. The response generation unit 606 generates a question sentence from the question answer record extracted from the question answer DB 548 by the question answer acquisition unit 602 and the input 54, and outputs it as a response (question) 570.

  For example, assume that the input 54 is “Like a Mona Lisa”. Assume that the response 566 of the dialogue system 52 is “There are always so many people around Mona Lisa that it seems difficult to see.” On the other hand, “Mona Lisa” is extracted as a topic. As a result, it is assumed that the question sentence of the question answer record extracted by the question answer acquisition unit 602 is “Do you know who drew Mona Lisa?”. In this case, for example, the response generation unit 606 connects the response 566 and the question sentence with a connection word such as “by the way”, and “It seems to be hard to see around Mona Lisa. “Do you know who drew Mona Lisa?” Is generated and output as a response (question) 570.

  In the selection unit 554, the determination signal 560 is a second value, and the determination signal 572 is a first value. When the value of the determination signal 560 is the second value and the value of the determination signal 572 is the first value, the selection unit 554 selects the response (question) 570. The input 54, this response (question) 570, and information for specifying the question response record used at that time are recorded in the dialogue history DB 544.

<When answering>
At the time of answering, information related to the immediately previous question is recorded in the dialogue history DB 544. Referring to FIGS. 15 and 18, when input 54 is given, answer determination unit 632 refers to dialog history DB 544 to determine whether or not the previous response from dialog system reinforcement device 530 is a question sentence. (Step 742 in FIG. 18 (hereinafter the same)). In this example, it is determined that the previous response was a question (YES in step 742). Furthermore, the answer determination unit 632 determines whether or not the input 54 is an answer to the question sentence (steps 744 and 746). In this example, the determination in step 746 is YES. Accordingly, the topic extraction unit 630 extracts words representing topics from the input 54 (step 748). The answer determination unit 632 extracts a question answer record from the question answer DB 548 using the topic extracted by the topic extraction unit 630 as a search key, and extracts an answer sentence (step 750). It is determined whether the topic of this answer sentence matches the topic of the immediately preceding question sentence (step 752). If they match, the determination in step 752 is YES. In step 754, a response at the time of correct answer is generated from the answer sentence, and output in step 762. If they do not match, a response at the time of an incorrect answer is generated from the answer sentence and output in step 762.

<Effect of Embodiment>
As described above, according to each of the above embodiments, the dialogue system can be converted into a dialogue system specialized in a certain category such as education and news without modifying the existing dialogue system. There is no need to modify the existing dialogue system or develop a new dialogue system. Therefore, the development cost can be reduced. In addition, the user basically interacts with an existing dialog system, and receives a specialized response only when a certain condition is satisfied. Therefore, it is possible to reduce the possibility of having a sense that the operation of the dialogue system has changed compared to when the existing dialogue system is used as it is. As a result, a specific new function can be added to the existing system without causing the user who used the existing dialog system to feel uncomfortable.

  In the second embodiment, the problem is adopted from the art and history, but the present invention is not limited to such an embodiment. It goes without saying that the dialogue system reinforcing apparatus of this embodiment is effective for any department that generally relates to education. In addition to education-related issues, it is also possible to provide so-called knowledgeable problems in dialogue. Further, in the above embodiment, the answer is also given in the form of a sentence. However, the present invention is not limited to such an embodiment. For example, it is possible to issue a so-called 3-choice, 4-choice problem, or a YES / NO problem.

<Realization by computer>
The functional units of the dialogue system reinforcement device 50 and the dialogue system reinforcement device 530 according to each of the above embodiments are respectively computer hardware, and a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit) on the hardware. And a program prepared as an add-on to the dialogue system. FIG. 19 and FIG. 20 show computer hardware that realizes each of the above apparatuses and systems. A GPU is usually used for performing image processing, and a technique for using a GPU for normal arithmetic processing instead of image processing in this way is called GPGPU (General-purpose computing on graphics processing units). The GPU can simultaneously execute a plurality of operations of the same type in parallel. On the other hand, at the time of training and operation of the neural network, the weight calculation of each node is a simple product-sum operation, and they can often be executed simultaneously. Accordingly, GPGPU is suitable for the training and operation of the MCNN 220 shown in FIGS. 4 to 6 and the MCNN (not shown) used for the score calculation of the question sentence shown in FIG. 17 and the score of the likelihood of answer shown in FIG.

  Referring to FIG. 19, the computer system 830 includes a computer 840 having a memory port 852 and a DVD (Digital Versatile Disk) drive 850, a keyboard 846, a mouse 848, and a monitor 842.

  Referring to FIG. 20, in addition to the memory port 852 and the DVD drive 850, the computer 840 includes a CPU 856 and a GPU 858, a bus 866 connected to the CPU 856, the GPU 858, the memory port 852 and the DVD drive 850, a boot program, and the like. It includes a ROM 860 that is a read-only memory for storing, a random access memory (RAM) 862 that is a computer-readable storage medium that is connected to the bus 866 and stores program instructions, system programs, work data, and the like, and a hard disk 854. The computer 840 is further connected to the bus 866, and inputs / outputs audio signals to / from a network interface (I / F) 844 that provides a connection to a network 868 that enables communication with other terminals. Voice I / F 870.

  A program for causing the computer system 830 to function as each functional unit of each device and system according to the above-described embodiments is a DVD 872 which is a computer-readable storage medium that is mounted on the DVD drive 850 or the memory port 852. Alternatively, it is stored in the removable memory 864 and further transferred to the hard disk 854. Alternatively, the program may be transmitted to the computer 840 through the network 868 and stored in the hard disk 854. The program is loaded into the RAM 862 when executed. The program may be loaded directly from the DVD 872 to the RAM 862 from the removable memory 864 or via the network 868. Data necessary for the above processing is stored in a predetermined address such as a hard disk 854, RAM 862, CPU 856 or a register in the GPU 858, processed by the CPU 856 or GPU 858, and stored in an address designated by the program. The MCNN parameters that have been finally trained are stored in, for example, the hard disk 854 or in the DVD 872 or the removable memory 864 via the DVD drive 850 and the memory port 852, respectively. Alternatively, the data is transmitted to another computer or storage device connected to the network 868 via the network I / F 844.

  This program includes an instruction sequence including a plurality of instructions for causing the computer 840 to function as each device and system according to the above-described embodiment. Numerical calculation processing in each of the above devices and systems is performed using the CPU 856 and the GPU 858. Only the CPU 856 may be used, but using the GPU 858 is faster. Some of the basic functions necessary to cause computer 840 to perform this operation are an operating system or third party program running on computer 840 or various dynamically linked programming toolkits or programs installed on computer 840. Provided by the library. Therefore, this program itself does not necessarily include all the functions necessary for realizing the system, apparatus, and method of this embodiment. The program is a system as described above, by dynamically calling the appropriate program in the appropriate function or programming toolkit or program library at run time in a controlled manner to obtain the desired result. It only needs to include instructions that implement the functions of the apparatus or method. Of course, all necessary functions may be provided only by the program.

  The embodiment disclosed herein is merely an example, and the present invention is not limited to the above-described embodiment. The scope of the present invention is indicated by each claim of the claims after taking into account the description of the detailed description of the invention, and all modifications within the meaning and scope equivalent to the wording described therein are included. Including.

50, 530 Dialog system reinforcement device 52 Dialog system 60, 600, 630 Topic extraction unit 62 News title acquisition unit 64, 604 Response availability determination unit 66 News DB
68,544 Dialog history DB
70, 74, 554 Selection unit 72, 606, 662 Response generation unit 78 News title 150 Model learning device 160 Large text database 162 Model learning unit 164 News term dictionary 166 Connection word dictionary 168 News likelihood determination model 170 News title storage unit 172 News title scoring unit 174 Scored news title 190 Text extraction / classification unit 192 Learning text storage unit 194 Learning processing unit 220 MCNN
260, 262, 264, 266 Convolutional neural network group 280 Subnetwork 360 News title candidate 540 Branch unit 542 State determination unit 548 Question answer DB
550 Question processing unit 574 Answer processing unit 602 Question answer acquisition unit 632 Answer determination unit 660 Answer sentence extraction unit

Claims (6)

A dialogue system reinforcement device that operates with a dialogue system and gives a feature to dialogue by the dialogue system,
First determination means for receiving an input given to the dialog system and determining whether or not to output a response to the input on behalf of the dialog system;
In response to a positive determination by the first determination unit, a first response generation unit for generating a response by a response generation method prepared in advance based on the input;
The output of the first response generation means is connected to receive the output of the first response generation means and the output of the dialogue system, and the output of the first response generation means is determined according to whether the determination result of the first determination means is affirmative. And a first selection means for selectively outputting the output of the dialogue system. The first determination means includes
A response sentence candidate storage unit storing a plurality of response sentence candidates so that a response sentence candidate related to the word can be searched using the word as a key;
Word extraction means for extracting words included in the input from the input;
Response sentence candidate extracting means for searching for and extracting response sentence candidates from the response sentence candidate storage means using the word extracted by the word extraction means as a key;
And means for determining whether or not to output a response to the input on behalf of the dialogue system according to whether or not there is a response sentence candidate extracted by the response sentence candidate extraction means. The dialogue system reinforcement device described. The first response generation means includes:
For each response sentence candidate extracted by the response sentence candidate extraction means, a score calculation means for calculating a feature score indicating the degree representing the feature;
Means for selecting a response sentence candidate having the highest feature score calculated by the score calculation means from among the response sentence candidates extracted by the response sentence candidate extraction means and outputting the response sentence candidate to the first selection means; The dialog system reinforcement apparatus of Claim 2 containing. The first response generation means includes:
A response sentence candidate storage unit storing a plurality of response sentence candidates so that a response sentence candidate related to the word can be searched using the word as a key;
Word extraction means for extracting words included in the input from the input;
Response sentence candidate extracting means for searching for and extracting response sentence candidates from the response sentence candidate storage means using the word extracted by the word extraction means as a key;
For each response sentence candidate extracted by the response sentence candidate extraction means, a score calculation means for calculating a feature score indicating the degree representing the feature;
Means for selecting a response sentence candidate having the highest feature score calculated by the score calculation means from among the response sentence candidates extracted by the response sentence candidate extraction means, and outputting the response sentence candidate to the first selection means; The dialog system reinforcement apparatus of Claim 1 containing this. The first response generation means includes:
A specific dialogue data item storage means for storing a plurality of specific dialogue data items;
Each of the specific dialogue data items includes a first response sentence, a second response sentence to be output after the first response sentence, and a first word included in at least the first response sentence. Including
The first response generation means further includes:
Word extraction means connected to the dialogue system for receiving a system response sentence output by the dialogue system in response to the input and for extracting a word included in the system response sentence;
Data item extraction means for extracting from the specific dialogue data item storage means a specific dialogue data item including the word extracted by the word extraction means as the first word;
Means for outputting the first response text included in the specific dialogue data item retrieved by the data item retrieving means and inputting it to the first selection means;
Second determination means for determining whether or not the specific dialog data item has been extracted by the data item extraction means;
History storage means for storing an output history indicating that the first response sentence has been output;
The first determination means refers to the history storage means in response to the input being given to the dialogue system, and is the input immediately after the output of the first response sentence? A third determining means for determining whether or not to output a response to the input on behalf of the dialogue system according to whether or not
The dialog system reinforcing device is further responsive to the second determination means determining that the input given to the dialog system is an input immediately after the output of the first response sentence. Second response generation means for generating and outputting a response based on the second response text in the same specific interaction data item as the first response text,
The first selection means is connected to receive the output of the first response generation means, the output of the dialogue system, and the output of the second response generation means, and the second determination means According to the combination of the determination result and the determination result by the third determination unit, the output of the first response generation unit, the output of the second response generation unit, and the output of the dialog system are selectively selected. The interactive system reinforcement device according to claim 1, further comprising second selection means for outputting. A computer program for causing a computer to function as each means of the interactive system reinforcement device according to any one of claims 1 to 5.

Images