JP2016189089A - Extraction equipment, extraction method and program thereof, support device, and display controller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide extraction equipment, a method and a program thereof that extract an abstract sentence on which the meaning of a sentence according to the order of words is reflected, and a support device, and a display controller.SOLUTION: An extraction device 10 comprises: a summary sentence extraction part 40 which extracts a plurality of sentences from a text including the plurality of sentences, generates a word set including one or more words for each of the sentences, calculates an evaluation value for evaluating the generated word set using a learning model, and extracts a summary sentence from the sentence on the basis of the calculated evaluation value. A display controller extracts the summary sentence from the sentence on the basis of the evaluation value calculated from the text including the plurality of sentences using the learning model, and performs display control over the summary sentence in order based upon the evaluation value.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an extraction device, an extraction method and a program thereof, a support device, and a display control device, and relates to extraction of a summary sentence from text.

  With the recent increase in demand for big data analysis, there is a growing need for teacher labels for machine learning for various information analysis purposes. The teacher label is information for instructing the machine learning device to which class the data belongs, or how much the score is when the machine learning device learns unknown data. However, the teacher label is not automatically determined by the apparatus, but it is necessary for a person to understand the contents of data and determine and assign the teacher label.

  Text data is information that expresses the personality and intention of the writer who wrote the text compared to numerical values or data such as images / videos (the vocabulary of the writer or the habit of word order / frequency of use and emotional expression) It is very useful data for analysis. However, unlike image / video data, text data cannot be understood at a glance, but it can be understood only by “reading”, so it takes a long time to understand the contents. In addition, since the work of “reading” greatly increases the work cost due to the complexity and length of the text, many techniques for supporting the understanding of the text content have been invented.

  The technique of Patent Document 1 specifies the genre of a document for a document to be read quickly, and selects a decision tree corresponding to the genre. On the other hand, features are extracted for each sentence in the body of a given document. The selected decision tree is compared with the characteristics of each sentence to determine whether each sentence is a summary sentence. The summary sentence is displayed in highlighted color and the non-summary sentence is displayed in background color. In addition, the first sentence of each paragraph of a given document is displayed in a color different from that of the summary sentence, thereby realizing fast reading support of the document by extracting and displaying important parts.

Japanese Patent No. 36552086

  In Patent Document 1, TF-IDF (Term Frequency-Inverse Document Frequency) representing characteristics based on the appearance frequency of words, or statistical characteristics of the number of characters is calculated, and each characteristic in the text is calculated based on comprehensive characteristics of the calculated words. It is judged whether the sentence seems to be a summary. For this reason, a summary sentence cannot be extracted reflecting the meaning of the words in the word order. For example, with respect to the word “important”, the meaning of the sentence may change greatly depending on whether the subsequent word is “is” or “not”, and a desired summary sentence cannot be extracted.

  The objective of this invention is providing the technique which can extract the summary sentence reflecting the meaning of the sentence accompanying the word order of a word.

  The extraction device of the present invention extracts the sentence from a text including a plurality of sentences, generates a word set including one or more words for each sentence, and evaluates the generated word set using a learning model. A summary sentence extracting unit that calculates an evaluation value to be extracted and extracts a summary sentence from the sentence based on the calculated evaluation value.

  The extraction method of the present invention extracts the sentence from text including a plurality of sentences, generates a word set including one or more words for each sentence, and evaluates the generated word set using a learning model. An evaluation value to be calculated is calculated, and a summary sentence is extracted from the sentence based on the calculated evaluation value.

  The extraction program of the present invention extracts a sentence from a text including a plurality of sentences to a computer, generates a word set including one or more words for each sentence, and applies a learning model to the generated word set. An evaluation value to be evaluated is calculated, and a summary sentence is extracted from the sentence based on the calculated evaluation value.

The support device of the present invention includes the above-described extraction device,
For each summary sentence output from the extraction device, the display is changed depending on whether or not it seems to be the teacher word set in the sentence.

  The display control apparatus of the present invention extracts a summary sentence from the sentence based on an evaluation value calculated from a text including a plurality of sentences using a learning model, and displays the summary sentence in an order based on the evaluation value The display control part to control is provided.

  The display control method of the present invention extracts a summary sentence from the sentence based on an evaluation value calculated from a text including a plurality of sentences using a learning model, and displays the summary sentence in an order based on the evaluation value. Control.

  The extraction device of the present invention can extract a summary sentence reflecting the meaning of words accompanying the word order.

It is a block diagram which shows the structure of the extraction apparatus by the 1st Embodiment of this invention. It is a flowchart which shows operation | movement of the extraction apparatus by the 1st Embodiment of this invention. It is a blow chart which shows operation | movement of the subroutine which calculates an evaluation value for every sentence shown in FIG. It is a figure explaining the example of the word N-Gram (N = 3) which connected three words. It is a figure for demonstrating calculation of a certainty factor. It is a blow chart which shows operation | movement of the subroutine of the summary sentence extraction shown in FIG. It is a figure for demonstrating the problem at the time of calculating a contribution by the sum total of certainty. It is a figure which shows the example which processes an English text by the word N-Gram (N = 3). It is a block diagram which shows the structure of the assistance apparatus and memory | storage device by the 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of the assistance apparatus by the 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of the subroutine of the language processing shown in FIG. It is a flowchart which shows operation | movement of the subroutine of the language processing shown in FIG. It is a figure which shows the screen display displayed on the display apparatus by 2nd Embodiment. It is a flowchart which shows operation | movement of the learning subroutine shown in FIG. It is a block diagram which shows the structure of the assistance apparatus and memory | storage device by 3rd Embodiment. It is a block diagram which shows the structure of the display control apparatus by the 4th Embodiment of this invention. It is a flowchart which shows operation | movement of the display control apparatus by the 4th Embodiment of this invention. It is a block diagram which shows the hardware constitutions which implement | achieved the extraction apparatus by the 1st Embodiment of this invention, the assistance apparatus by the 2nd, 3rd embodiment, or the display control apparatus by 3rd Embodiment with the computer apparatus.

<First Embodiment>
An extraction apparatus according to a first embodiment of the present invention will be described with reference to the drawings. The extraction device 10 according to the first embodiment is a device that provides one function of a support device that supports determination of a teacher label to a person who gives a teacher label to a text.

  FIG. 1 is a block diagram illustrating a configuration of an extraction device 10 according to the first embodiment. As illustrated in FIG. 1, the extraction device 10 includes a summary sentence extraction unit 40. The summary sentence extraction unit 40 extracts a sentence from a text including a plurality of sentences, generates a word set including one or more words for each sentence, and evaluates the generated word set using a learning model. And a summary sentence is extracted from the sentence based on the calculated evaluation value.

  As an example of extracting a sentence from a text including a plurality of sentences, the summary sentence extracting unit 40 divides the target text, which is a text to which a teacher label is attached, into words separated by words constituting the target text. Is divided into sentence units. Furthermore, as an example of calculating an evaluation value to be evaluated using a learning model for the generated word set, the summary sentence extraction unit 40 includes a word N-Gram in which N words are connected to each divided sentence. (N is a natural number of 2 or more). Subsequently, the summary sentence extraction unit 40 calculates a certainty factor representing the likelihood of the teacher label using the learning model for the generated word N-Gram. Further, as an example of extracting a summary sentence from a sentence based on the calculated evaluation value, the summary sentence extraction unit 40 calculates a teacher label determination contribution for each sentence divided based on the calculated certainty factor. Then, a summary sentence is extracted according to the teacher label determination contribution. The teacher label determination contribution will be described later.

  Next, the operation of the extraction device according to the first embodiment of the present invention will be described with reference to the drawings. FIG. 2 is a flowchart showing the operation of the extraction device 10 according to the first embodiment. As illustrated in FIG. 2, the extraction device 10 extracts a sentence from a text including a plurality of sentences (S101). Specifically, the extraction apparatus 10 divides the target text, which is a text segmented by words constituting the target text, into the target text that is the text to which the teacher label is attached.

Here, the word-delimited text means text that is expressed in split text. For example, if the Japanese text is “The shop is far from home, but the food is very delicious.” For each word, “The shop is far from home, but the food is very delicious.” This is delimited text. Further, to divide into sentences means to divide sentences according to the existence and position of punctuation marks (such as punctuation marks) added in the text. For example, the sentence “The shop is far from home but the food is very delicious” in the previous example is “The shop is far from home,” depending on the position of the reading, “The food is very delicious.” Divided into two sentences. Note that the division into sentence units can be processed in the following units in addition to the positions of the punctuation objects.
・ "Unit of word M-Gram (M> N: M, N is a natural number of 2 or more) with more word breaks than word N-Gram"
・ "Every K characters (K is a natural number of 1 or more)"
・ "Line unit (line feed character)"
・ "Page unit (page break code)"
・ "Section, paragraph unit"
Next, the extraction apparatus 10 generates a word set including one or more words for each sentence, and calculates an evaluation value for evaluating the generated word set using a learning model (S102). Specifically, a certainty factor representing the likelihood of a teacher label is calculated, and a teacher label determination contribution (hereinafter referred to as a contribution) is calculated for each sentence divided based on the calculated certainty factor. The teacher label determination contribution is a value indicating the degree of contribution to the determination of the person who gives the teacher label.

  FIG. 3 is a flowchart showing the operation of a subroutine for calculating an evaluation value for each sentence. As illustrated in FIG. 3, the extraction apparatus 10 generates a word set for each sentence (S1021). Specifically, the extraction device 10 generates a word N-Gram (N is a natural number of 2 or more) in which N words are connected for each divided sentence.

  FIG. 4 is a diagram illustrating an example of a word N-Gram (N = 3) in which three words are connected. As shown in Fig. 4, one sentence separated by words, "The shop I booked last week was very popular" was converted into a string containing three consecutive words for each word. is there. In the example of FIG. 4, ten words N-Gram (N = 3) are generated.

Here, the certainty factor, which is an example of an evaluation value, is a teacher label score calculated for each generated word N-Gram. Therefore, the score of the teacher label of the word N-Gram (N = 3) is a score indicating the degree of whether a character string including three consecutive words is P (positive) or N (negative). is there.
The learning model is constructed as follows. First, text for which P / N information (positive / negative information) is known is used as teaching data for learning. Subsequently, a learning unit (not shown) that generates a learning model creates a word N-Gram of text serving as teacher data, and then replaces each word N-Gram with a feature vector associated with the word, thereby learning model A feature vector and P / N information (score) are taught. This gives the learning model the ability to determine P / N information (score) from the text. The learning model can be generated using any supervised machine learning classifier, such as, for example, a support vector machine, a neural network, or a Bayes classifier. In the first embodiment, it is assumed that a learning model for calculating a certainty factor is prepared in advance before calculating the certainty factor. It can be said that the learning model is a model learned by using a plurality of teacher word sets so that it can be evaluated whether or not the predetermined word set seems to be a teacher word set.

  Next, the extraction apparatus 10 calculates an evaluation value for evaluating the generated word set using a learning model (S1022).

  FIG. 5 is a diagram for explaining calculation of certainty by the extraction device 10 using a learning model generated using a neural network and each generated word N-Gram. The extraction device 10 calculates a certainty factor for each generated word N-Gram as a test data (target text to which a teacher label is assigned) with unknown P / N information using a learning model. The example shown in FIG. 4 is used as the target text to which the teacher label is assigned.

  As illustrated in FIG. 5, the extraction device 10 replaces each generated word N-Gram (N = 3) with a feature vector associated with the word. Next, the extraction device 10 inputs a feature vector associated with the replaced word for each word N-Gram (N = 3) to a learning model generated using a neural network. Subsequently, the extraction apparatus 10 estimates a score of P / N (positive / negative) information for each word N-Gram (N = 3). The certainty score range is from -1 to 1 (0.1 unit). N-grams that are likely to appear in both P (positive) and N (negative) are near “0”, N-grams that are likely to appear in positive sentences are N-grams that are likely to appear in negative sentences near “1”. Gram is set to be near “−1”. In the example of FIG. 5, the certainty factor (teacher label score) is calculated for every ten words N-Gram (N = 3).

The extraction device 10 extracts a summary sentence from the sentence based on the calculated evaluation value (S103). FIG. 6 is a flowchart showing the operation of a subroutine for extracting a summary sentence. As illustrated in FIG. 6, the extraction device 10 calculates a teacher label determination contribution for each sentence based on the calculated evaluation value (S1031). Specifically, the extraction device 10 calculates a contribution for each divided sentence based on the certainty factor calculated for each word N-Gram.
As an example of the calculation of the contribution degree by the extraction device 10, the following variations can be considered.
-The variance value or standard deviation value of the certainty factor calculated for each word N-Gram-The maximum absolute value of the certainty factor calculated for each word N-Gram-The norm value of the certainty factor calculated for each word N-Gram -Average value of calculated certainty factor in word N-Gram Here, a problem caused by the total sum of calculated certainty factors will be described. FIG. 7 is a diagram illustrating an example of values calculated by the sum of certainty factors. As shown in FIG. 7, the upper part represents positive / negative (P / N) in the certainty factor for each of the calculated eight words N-Gram (N = 3), and the lower part represents the score. When the contribution for each sentence is calculated using only the sum of the certainty based on the certainty shown in FIG. 7, the total value of the sum is 0.00. That is, as shown in FIG. 6, although the positive / negative value as the certainty is an extremely large numerical value, the contribution for each sentence becomes 0.00 due to the sum, so that the summary sentence in the latter stage is extracted. In this case, the sentence may not be extracted as an important summary sentence.

  To solve this problem, in the first embodiment, a variance value or a standard deviation of the calculated certainty factor in each word N-Gram is used for calculating the contribution. As a result, as shown in FIG. 7, the variance value becomes 0.9 and the standard deviation value becomes 0.95, and an important summary sentence can be extracted.

  In addition, by adopting the “maximum absolute value of certainty” as a contribution, summary sentence extraction including one or more words N-Gram having high certainty (estimated by the learning model with confidence) is also possible. I can think of it. It is also possible to extract a summary sentence by combining these variations.

  Finally, the extraction device 10 extracts a summary sentence from the contribution calculated for each sentence (S1032). As a summary sentence extraction condition, a sentence whose calculated contribution is equal to or greater than a predetermined threshold or a sentence that is the top tens percent of the calculated contributions arranged in descending order is extracted. The above summary sentence extraction conditions are only examples, and other extraction conditions are also applicable.

  Moreover, although 1st Embodiment showed the example of Japanese as a text which assign | provides a teacher label, it is not restricted to this, It can apply also to English object text. FIG. 8 is a diagram illustrating an example of processing an English target text with a word N-Gram (N = 3). In the case of a target text that is usually divided, such as English, there is no need to divide the target text into words. As illustrated in FIG. 8, the extraction device 10 calculates the certainty factor for each word N-Gram (N = 3) generated for each sentence, and based on the calculated certainty factor, the teacher label determination contribution for each sentence. Calculate the degree. As a result, even in English text, a summary sentence reflecting the meaning of words accompanying the word order can be extracted.

  According to the extraction apparatus of the first embodiment, it is possible to extract a summary sentence that reflects the meaning of the word accompanying the word order. For example, in the text “The shop is clean and the atmosphere is not bad” and the text “The atmosphere is bad and the shop is not clean”, the words that make up the text are the same. For this reason, in an example in which the word extraction is performed in units of words as in Patent Document 1 and the word appearance frequency is used, when the meaning changes depending on the combination of words, the meaning of the order cannot be taken into consideration, and the desired summary The sentence cannot be extracted. On the other hand, according to the extraction device 10 according to the first embodiment, since the calculation is performed for each N-Gram such as “ha is beautiful” and “is not bad”, even when the sentence changes depending on the combination of words. A desired summary sentence can be extracted. That is, it is possible to extract a summary sentence that reflects the meaning of the word order.

  Also, in the example where only sentence units are extracted, if there are a plurality of sentences in one sentence (for example, there are a plurality of P (positive) and N (negative) in one sentence as shown in FIG. 7), A summary sentence cannot be extracted. On the other hand, the extraction device 10 according to the first embodiment calculates the contribution degree by calculating the variance value or standard deviation value of the certainty factor calculated in each word N-Gram, the maximum absolute value of the certainty factor calculated, Or it uses with the norm value of the calculated certainty factor. This makes it possible to extract an appropriate summary sentence even if there are a plurality of meanings in one sentence.

<Second Embodiment>
A support apparatus according to a second embodiment of the present invention will be described with reference to FIG. FIG. 9 is a block diagram illustrating a configuration of the support device according to the second embodiment. The support device 1 is connected to a display device 5 and a storage device 6.

  The support device 1 has a function of extracting a summary sentence from a text to which a teacher label is attached using a text (target text) to which a teacher label is attached and a machine learning model. Further, the support device 1 may have a function of presenting the extracted summary sentence to the user of the support system via the display device 5 connected to the support device 1. Specifically, the display control unit (not shown) controls display of the extracted summary sentence. Further, the target text is acquired via a communication unit (not shown) of the support device 1.

  The storage device 6 has a function of storing various data for the support device 1 to acquire, generate, or calculate.

  The display device 5 has a function of displaying the information of the summary sentence that is extracted from the support device 1 and is extracted to give the teacher label.

  The support device 1 and the storage device 6 according to the second embodiment will be described in detail with reference to the drawings.

  The support device 1 includes an extraction device 10, a language processing unit 20, a learning unit 30, and a teacher label receiving unit 50. Further, the extraction device 10 includes a summary sentence extraction unit 40. In the description of the support device of the second embodiment, the same reference numerals are given to the same components as those of the first embodiment, and the description is simplified.

  The language processing unit 20 of the support apparatus 1 acquires the target text that is the text to which the teacher data is added, divides the acquired target text into words, and converts the words constituting the target text and the word-delimited text data. It has a function to generate. The language processing unit 20 passes the generated word break target text to the summary sentence extraction unit 40 of the extraction device 10 or stores it in the text storage unit 62 of the storage device 6.

  The learning unit 30 of the support device 1 acquires the target text of the word break and indexes it with the words constituting the target text stored in the word storage unit 61. Further, after creating a feature vector for each word N-Gram, the learning unit 30 learns the learning model stored in the learning model storage unit 63 along the parameters read from the parameter storage unit 64. Here, the parameters are teacher data (text with known P / N information, P / N information (score), etc.) used to create a learning model. Or may be acquired from the text storage unit 62 of the storage device 6.

  The teacher label receiving unit 50 of the support device 1 receives the teacher label determined by the user of the support system for the text to which the teacher label is assigned, and stores the determined teacher label result in the text storage unit 62. . A general input device can be applied to accept the determined teacher label. For example, a mouse, a keyboard, a touch panel, or the like can be used.

  Next, the configuration of the storage device 6 connected to the support device 1 will be described with reference to FIG. The storage device 6 includes a word storage unit 61, a text storage unit 62, a learning model storage unit 63, and a parameter storage unit 64.

  The word storage unit 61 of the storage device 6 stores words constituting the target text input to the support device 1.

  The text storage unit 62 of the storage device 6 stores the target text or the word break target text input to the support device 1 and the teacher label paired with the target text.

  The learning model storage unit 63 of the storage device 6 stores a learning model for learning the target text input to the support device 1.

  The parameter storage unit 64 of the storage device 6 stores parameters used for creating and learning a learning model.

  Although the storage device 6 is arranged outside the support device 1 and connected to the support device 1, the storage device 6 is placed inside the support device 1 and connected to the support device 1. Also good.

  Next, operation | movement of the assistance apparatus 1 by the 2nd Embodiment of this invention is demonstrated using drawing. FIG. 10 is a flowchart showing the operation of the support apparatus 1 according to the second embodiment of the present invention.

As illustrated in FIG. 10, the support apparatus 1 acquires text (target text) to which a teacher label is attached. The language processing unit 20 of the support device 1 divides the acquired target text into word units, which are word-delimited target text divided by words constituting the target text (S201).
FIG. 11 is a flowchart showing the operation of a subroutine for language processing (S201).
The language processing unit 20 performs morphological analysis on the acquired target text and divides the target text into word breaks (S2011). The language processing unit 20 sends the divided words and the target text of word breaks to the summary sentence extraction unit 40, respectively. Instead of being sent to the summary sentence extraction unit 40, the divided words and the target text of word breaks may be temporarily stored in a storage device (not shown).

  FIG. 12 is a flowchart showing another example of the operation of the subroutine of language processing (S201). In the operation of the subroutine shown in FIG. 12, the language processing unit 20 performs a syntax analysis for analyzing the dependency of the morpheme (S2012) after the morphological analysis (S2011) of FIG. The language processing unit 20 performs syntax analysis to obtain word dependency information, and can provide additional information when calculating the certainty factor of the word N-Gram in the extraction device 10 in the subsequent stage. The appropriateness of summary extraction will be improved.

  In the language processing step (S201), an example in which morphological analysis is used for word separation of the target text has been shown. However, as long as word separation target text can be generated, other than morphological analysis may be used. Further, in the case of a target text that is preliminarily written like English text, a process for dividing the target text into sentences may be performed.

  The divided words are stored in the word storage unit 61, and the target text that is a word break that is a sentence extracted from text including a plurality of sentences is stored in the text storage unit 62. Alternatively, it is sent by the language processing unit 20 to the summary sentence extraction unit 40 of the subsequent extraction device 10.

  Next, the extraction device 10 of the support device 1 generates a word set including one or more words for each sentence, calculates an evaluation value to be evaluated using a learning model for the generated word set, and is calculated A summary sentence is extracted from the sentence based on the evaluation value (S202). Specifically, the extraction apparatus 10 generates a word N-Gram (N is a natural number of 2 or more) in which N words are connected for each divided sentence, and a learning model is generated for the generated word N-Gram. Is used to calculate a certainty factor representing the likelihood of a teacher label. Subsequently, the extraction device 10 calculates a teacher label determination contribution for each sentence divided based on the calculated certainty factor, and extracts a summary sentence according to the teacher label determination contribution. Further, the extraction device 10 outputs the extracted summary sentence to the display device 5.

  The summary sentence extraction step by the extraction device 10 is the same as the operation of the extraction device 10 of the first embodiment, and thus detailed description thereof is omitted. Note that the word break target text generated by the language processing unit 20 may be acquired from the text storage unit 62 of the storage device 6 or may be acquired from the language processing unit 20 for extracting the summary sentence. .

  Next, the display device 5 displays the summary sentence output from the extraction device 10 of the support device 1 (S203). FIG. 13 is a diagram illustrating a screen display output to the display device 5 by the support device 1 according to the second embodiment. As illustrated in FIG. 13, the display screen of the display device 5 includes four display areas of “text list”, “option”, “teacher label”, and “text”.

  The display area of “text list” displays a list of target texts (texts to which teacher labels are attached). The target text to be displayed in the text list may be in the order input to the support device 1 or in a predetermined descending order. The user of the support system selects the target text from those displayed in the “text list” display area.

  The “text” display area is an area for displaying the text selected in the “text list”. The display control unit changes the display for each summary sentence output from the extraction device 10 according to whether or not it seems to be a teacher word set in the sentence. In FIG. 13, for each summary sentence extracted, an underline is drawn in a portion estimated to be “positive” in the sentence.

  The “option” display area is an area for selecting the summary text display mode or the full text display mode for the target text, and the content displayed in the “text” display area is changed according to the selection state. This option is prepared in order to avoid the problem when the extraction of the summary sentence is not valid in the initial stage of the support system in which machine learning is insufficient. In other words, it is possible to operate the support system efficiently in assigning teacher labels by displaying the entire text in the initial stage of operation of the support system and selecting to display a summary sentence when the teacher labels are sufficient. Become.

  The “teacher label” display area is an area for setting / changing the teacher label for the target text displayed in “text”. In FIG. 13, the teacher label is selected as “positive”.

  The support device 1 displays the extracted summary sentence on the display device 5 so that the user of the support device 1 can understand the content of the text in a short time, thereby reducing the time and effort for determining the teacher label. It becomes possible to do.

  Next, the teacher label receiving unit 50 of the support apparatus 1 receives the result of the teacher label determined by the user of the support system (S204). The teacher label receiving unit 50 stores the teacher label determined by the user of the support system in the text storage unit 62 in association with the target text.

  Subsequently, the learning unit 30 of the support device 1 learns the learning model stored in the learning model storage unit 63 using the teacher label received by the teacher label receiving unit and the target text associated therewith (S205). .

  FIG. 14 is a flowchart showing the operation of the learning subroutine shown in FIG. The learning referred to here includes a feature vector creation process for learning the target text (since there is an algorithm in which the feature vector creation process is integrated with the learning process).

First, the learning unit 30 creates a vector used for learning (S2051). In general, a feature vector of a natural language is data having a very large vector length, and if applied as it is, it is difficult to apply to subsequent learning and discrimination. For this reason, only a term that becomes a feature is selected and a compressed vector is generated. For example, the generation of feature vectors is described in detail in the following paper.
“Sentiment Classification with Supervised Sequence Embedding”, Bespalov, Dmitriy and Qi, Yanjun and Bai, Bing and Shokoufandeh, Ali, Machine Learning and Knowledge Discovery in Databases, Vol.7523, pp.159-174, Springer Berlin Heidelberg, 2012, ISBN : 978-3-642-33459-7
The above paper uses a mechanism that automatically processes the generation of feature vectors. In the second embodiment, the present invention is not limited to this. For example, an important vector term is analyzed by principal component analysis, the vector term is selected, and a process for generating a feature vector is incorporated in the software program. May be.

  Subsequently, the learning unit 30 reads the learning model from the learning model storage unit 63 and corrects the learning model using the vector created in the vector creation step (S2051). An arbitrary supervised machine learning classifier can be applied to the learning model adopted in the learning model storage unit 63. In addition, a support vector machine, a neural network, a Bayes classifier, or the like may be used.

<Third Embodiment>
A support device and a storage device according to a third embodiment of the present invention will be described with reference to FIG. FIG. 15 is a block diagram illustrating configurations of the support device 1 and the storage device 7 according to the third embodiment. As illustrated in FIG. 15, the storage device 7 of the third embodiment is different from the storage device 6 of the second embodiment in that an additional information storage unit 65 is added.

  The additional information storage unit 65 of the storage device 7 according to the third embodiment stores attribute information related to the target text. Thereby, the support device 1 according to the third embodiment can learn using the attribute information of the target text. In the example of the second embodiment, learning is performed by handing over the words constituting the target text stored in the word storage unit 61 and the text storage unit 62 and the data of the word break target text to the learning unit 30. In addition, in the example of the third embodiment, additional information such as the genre of the target text (article, novel, etc.), the author's domain (gender, age, etc.), and the layout (sentences of the sentence as seen in the entire text, number of characters). To learn. Thereby, the precision which extracts a summary sentence improves.

<Fourth Embodiment>
The display control apparatus which is the 4th Embodiment of this invention is demonstrated using drawing. FIG. 16 is a block diagram illustrating a configuration of the display control apparatus 110 according to the fourth embodiment. The display control apparatus 110 according to the fifth embodiment is a display control apparatus for assisting a person who gives a teacher label to a text to determine the teacher label.

  As shown in FIG. 16, the display control device 110 includes a display control unit 140. The display control unit 140 extracts a summary sentence from a sentence based on an evaluation value calculated using a learning model from a text including a plurality of sentences, and controls display of the summary sentence in an order based on the evaluation value. A summary sentence is extracted from a sentence based on an evaluation value calculated using a learning model from a text including a plurality of sentences, similar to the extraction device 10 according to the first embodiment.

  FIG. 17 is a flowchart showing the operation of the display control apparatus 110 according to the fourth embodiment. As shown in FIG. 17, the display control apparatus 110 extracts a summary sentence from a sentence based on an evaluation value calculated using a learning model from a text including a plurality of sentences (S111), and uses the summary sentence as an evaluation value. A sentence is extracted from a text including a plurality of sentences whose display is controlled in the order based on the order (S112). In addition, if the teacher label determination contribution is calculated for each sentence based on the evaluation value calculated using the learning model from the text including multiple sentences, and the summary sentence is extracted according to the contribution degree, the summary sentence contributes Display control may be performed in the order based on the degree.

  According to the fourth embodiment, it is possible to control display of a summary sentence reflecting the meaning of words accompanying the word order. Thereby, it becomes possible to assist the person who gives the teacher label to the text to determine the teacher label.

(Hardware configuration)
FIG. 18 shows a hardware configuration in which the extraction device 10 according to the first embodiment of the present invention, the support device 1 according to the second and third embodiments, or the display control device 110 according to the fourth embodiment is realized by a computer device. FIG.

  The computer apparatus shown in FIG. 18 includes a CPU (Central Processing Unit) 91, a network connection communication I / F (communication interface) 92, a memory 93, and a storage device 94 such as a hard disk for storing a program. The computer device is connected to an input device 95 and an output device 96 via a system bus 97.

  The CPU 91 operates the operating system, and the summary sentence extraction unit 40 of the extraction device 10 according to the first embodiment, the language processing unit 20 of the support device 1 according to the second embodiment, the learning unit 30, and the teacher label reception unit 50. Alternatively, the display control unit 140 of the display control apparatus according to the fourth embodiment is controlled. Further, the CPU 91 reads out programs and data from the recording medium mounted on the drive device to the memory 93, for example. Further, for example, the CPU 91 has a function of processing an information signal in each embodiment, and executes processing of various functions based on a program.

  The storage device 94 is, for example, an optical disk, a flexible disk, a magnetic optical disk, an external hard disk, or a semiconductor memory. A part of the storage medium of the storage device 94 is a nonvolatile storage device, and stores a program therein. The program is connected to a communication network. It may be downloaded from an external computer (not shown).

  The input device 95 is realized by, for example, a mouse, a keyboard, a built-in key button, a card slot, or a touch panel, and is used for an input operation.

  The output device 96 is realized by a display, for example, and is used for outputting and confirming information processed by the CPU 91.

  As described above, each embodiment of the present invention is realized by the hardware configuration shown in FIG. However, the means for realizing each unit included in the extraction device 10 or the support device 1 is not particularly limited. That is, the extraction device 10 or the support device 1 may be realized by one physically coupled device, or two or more physically separated devices are connected by wire or wirelessly, You may implement | achieve with an apparatus.

  While the present invention has been described with reference to the embodiments (and examples), the present invention is not limited to the above embodiments (and examples). Various changes that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

  Part or all of the above-described embodiments can be described as in the following supplementary notes, but is not limited thereto.

(Appendix 1)
A word N is obtained by dividing a target text, which is a text to be assigned a teacher label, into word units divided by words constituting the target text into sentence units and connecting N words for each of the divided sentences. -Gram (N is a natural number greater than or equal to 2) is generated, a certainty factor representing the likelihood of a teacher label is calculated using a learning model for the generated word N-Gram, and the above-described certainty factor An extraction apparatus including a summary sentence extraction unit that calculates a teacher label determination contribution for each divided sentence and extracts a summary sentence according to the contribution.

(Appendix 2)
The summary sentence extraction unit includes:
Unit of word M-Gram (M> N: M, N is a natural number of 2 or more), K characters (K is a natural number of 1 or more), line unit (line feed character) having more word breaks than the word N-Gram Divide the target text of the word break into sentence units by page unit (page break code), punctuation unit, or section / paragraph unit,
The extraction device according to attachment 1.

(Appendix 3)
The learning model is
A word N-Gram using text whose score information is known teacher data is created, replaced with a feature vector associated with the word for each created word N-Gram, and the score information corresponding to the feature vector Is a model trained by any supervised machine learning classifier with
The extraction apparatus according to Supplementary Note 1 or Supplementary Note 2.

(Appendix 4)
The learning model is any supervised machine learning classifier, either a support vector machine, a neural network, or a Bayes classifier.
The extraction device according to any one of supplementary notes 1 to 3.

(Appendix 5)
The calculation of the contribution is performed by calculating the variance value or standard deviation value of the certainty factor calculated in each word N-Gram, the maximum absolute value of the certainty factor calculated in each word N-Gram, or in each word N-Gram. Use one of the calculated norm values of confidence,
The extraction device according to any one of supplementary notes 1 to 4.

(Appendix 6)
The summary sentence extracts a sentence in which the calculated contribution is equal to or greater than a predetermined threshold, or a sentence that is the top tens percent of the calculated contributions arranged in descending order.
The extraction device according to any one of appendices 1 to 5.

(Appendix 7)
Including the extraction device,
The support device according to any one of supplementary notes 1 to 6.

(Appendix 8)
A language processor,
The language processing unit generates the target text of the word break;
The support apparatus according to appendix 7.

(Appendix 9)
With a learning unit,
The learning unit creates a word N-Gram using text that is teacher data with known score information, replaces the created word N-Gram with a feature vector associated with the word, Let any supervised machine learning classifier learn with the corresponding score information,
The support apparatus according to appendix 7 or appendix 8.

(Appendix 10)
Equipped with a teacher label reception,
The teacher label receiving unit receives a teacher label determined by a user of the support device for the target text;
The support device according to any one of appendix 7 to appendix 9.

(Appendix 11)
A storage device is connected to the support device;
The storage device includes a word storage unit, a text storage unit, a learning model storage unit, and a parameter storage unit.
The support device according to any one of appendix 7 to appendix 10.

(Appendix 12)
The storage device includes an additional information storage unit.
The support device according to attachment 11.

(Appendix 13)
Comprising the storage device;
The support device according to appendix 11 or appendix 12.

(Appendix 14)
A display device is connected to the support device;
The display device
The support device according to any one of appendix 7 to appendix 13.

(Appendix 15)
Comprising the display device;
The support device according to any one of appendix 7 to appendix 14.

(Appendix 16)
Dividing the target text of the word break, which is the target text that is the text to which the teacher label is attached, separated by the words constituting the target text into sentence units,
Generating a word N-Gram (N is a natural number of 2 or more) by connecting N words for each of the divided sentences;
Calculating a certainty factor representing the likelihood of a teacher label using a learning model for the generated word N-Gram;
Calculating a teacher label determination contribution for each of the divided sentences based on the calculated certainty factor, and extracting a summary sentence according to the contribution degree;
Extraction method.

(Appendix 17)
On the computer,
Dividing the target text of the word break, which is the target text that is the text to which the teacher label is attached, separated by the words constituting the target text into sentence units,
Generating a word N-Gram (N is a natural number of 2 or more) by connecting N words for each of the divided sentences;
Calculating a certainty factor representing the likelihood of a teacher label using a learning model for the generated word N-Gram;
Calculating a teacher label determination contribution for each of the divided sentences based on the calculated certainty factor, and extracting a summary sentence according to the contribution degree;
Extraction program to make things happen.

DESCRIPTION OF SYMBOLS 1 Support apparatus 5 Display apparatus 6 Storage apparatus 7 Storage apparatus 10 Extraction apparatus 20 Language processing part 30 Learning part 40 Summary sentence extraction part 50 Teacher label reception part 61 Word storage part 62 Text storage part 63 Learning model storage part 64 Parameter storage part 65 Additional information storage unit 91 CPU
92 Communication I / F (communication interface)
93 Memory 94 Storage Device 95 Input Device 96 Output Device 97 System Bus 110 Display Control Device 140 Display Control Unit

Claims (10)

  Extracting the sentence from a text including a plurality of sentences, generating a word set including one or more words for each sentence, calculating an evaluation value to be evaluated using a learning model for the generated word set; An extraction apparatus comprising a summary sentence extraction unit that extracts a summary sentence from the sentence based on the calculated evaluation value. The learning model is a model learned using a plurality of teacher word sets so that it can be evaluated whether or not a predetermined word set seems to be the teacher word set.
The extraction device according to claim 1. The summary sentence extraction unit calculates a teacher label determination contribution for each sentence based on the calculated evaluation value, and extracts a summary sentence according to the contribution;
The extraction device according to claim 1 or 2. The calculation of the contribution degree uses either a variance value or a standard deviation value of the calculated evaluation value, a maximum absolute value of the calculated evaluation value, or a norm value of the calculated evaluation value.
The extraction device according to claim 3.   Extracting the sentence from a text including a plurality of sentences, generating a word set including one or more words for each sentence, calculating an evaluation value to be evaluated using a learning model for the generated word set; An extraction method for extracting a summary sentence from the sentence based on the calculated evaluation value.   Extracting the sentence from a text including a plurality of sentences, generating a word set including one or more words for each sentence, calculating an evaluation value to be evaluated using a learning model for the generated word set; An extraction program for causing a computer to extract a summary sentence from the sentence based on the calculated evaluation value. An extraction device according to any one of claims 1 to 4,
A support device that changes a display for each summary sentence output from the extraction device according to whether or not it seems to be a teacher word set in the sentence.   A display including a display control unit that extracts a summary sentence from the sentence based on an evaluation value calculated from a text including a plurality of sentences using a learning model, and controls display of the summary sentence in an order based on the evaluation value Control device.   A display control method for extracting a summary sentence from the sentence based on an evaluation value calculated from a text including a plurality of sentences using a learning model, and controlling the display of the summary sentence in an order based on the evaluation value. )
A teacher label determination contribution is calculated for each sentence based on an evaluation value calculated from a text including a plurality of sentences using a learning model, a summary sentence is extracted according to the contribution degree, and the summary sentence is A display control method for performing display control in an order based on the contribution degree.

Images