JP6059598B2 - Information extraction method, information extraction apparatus, and information extraction program - Google Patents

Description

  The present invention relates to a technique for retrieving information from a document.

  By automatically extracting event information from web pages and blog articles that describe local event information, it is possible to construct an event information database without human labor, and use it for event recommendation services, etc. can do.

  In order to extract the event information from the text, for example, using non-patent documents 1 to 3, candidates for each category such as event name, place, date and time can be extracted. In addition, if there is correct data tagged manually for each category, it is possible to construct a discriminator that automatically discriminates against the event name, location, date and time using a supervised machine learning framework. Event name, location, date and time can be automatically extracted from web documents.

Yamada, et al., “Japanese Named Expression Extraction Using Support Vector Machine”, Transactions of Information Processing Society of Japan, Information Processing Society of Japan, January 2002, Vol. 43, No. 1, pp.44-53 Hirano and two others, “Resolving ambiguity of place names using geographical distance and famousness”, 70th Information Processing Society of Japan Conference, Information Processing Society of Japan, 2008, pp.2-85-2-86 Takashima and three others, “Specified date and time search considering the valid range of written date and time”, 3rd Forum on Web and Database, 2010 Hira, et al., “Predicate term structure analysis using structural learning”, Proc. Of the 14th Annual Conference of the Language Processing Society of Japan, 2008, pp.556-559 Crammer K., et al., “Online Passive-Aggressive Algorithms”, Journal of Machine Learning, 2006, Vol. 7, pp. 551-585

  Instead of individually applying a discriminator, there is a method using structure output learning in which a structure is provided for prediction and a model for selecting a correct combination from extracted candidates is constructed. In this case, for example, a prediction model can be constructed using Non-Patent Document 5.

  However, since a method using specific expression extraction such as Non-Patent Document 1 is not an event name extraction method, there is a problem that acquisition of an appropriate event name candidate fails. Since the prediction model of the present invention selects a combination from given prediction candidates, if there is no appropriate candidate, an appropriate event name cannot be acquired, and the accuracy may be reduced.

  The present invention has been made in view of the above, and an object of the present invention is to prevent an increase in learning cost while extracting appropriate candidates when extracting information of relevant categories from a document.

The information extraction apparatus according to the first aspect of the present invention includes an extraction candidate storage unit that extracts and stores candidates for each related category from the document information, and among the candidates extracted from the document information, the candidates in the same category If the ratio of the number of characters of the candidate with respect to the number of characters of the other candidate is less than a preset removal ratio, a candidate removing unit that removes the candidate, and a category Correct answer storage means for storing each correct answer, extraction model storage means for extracting an extraction model for extracting information of relevant categories, and all candidates for each category stored in the extraction candidate storage means for the possible combinations, the feature vector calculating means for calculating a feature vector indicating the feature of the combination were stored in the extraction model storage means A combination acquisition unit that acquires a combination having a maximum score calculated using the outgoing model and the feature vector; and a calculation of a score by reading the combination of the correct answers from the correct answer storage unit, and a score for the combination of the correct answers An extraction model update unit that updates the extraction model stored in the extraction model storage unit when the loss of the combination score acquired by the combination acquisition unit is within a predetermined range; a prediction target extracting candidate storage means for storing extracted from the prediction target document information, for all possible combinations of the prediction target extraction candidate storage unit the category of each stored in a candidate, indicating the feature of the combination A feature vector is calculated, and the extracted model stored in the extracted model storage means and the feature vector are stored. Score is calculated using the torque is characterized by having a, an information extracting means for obtaining a combination that maximizes.

  The information extraction apparatus further includes cost storage means for storing an error cost when the candidate is the correct partial character string and when the candidate is not the partial character string, and the extraction model update means includes the cost storage means from the cost storage means. An error cost is read out, a cost corresponding to the proportion of the candidates corresponding to the correct answer is obtained, and the obtained cost is set as the predetermined range.

An information extraction method according to a second aspect of the present invention is an information extraction method executed by a computer, wherein the candidates stored in extraction candidate storage means for extracting and storing candidates for each related category from document information are stored. If the candidate is a partial character string of another candidate of the same category, if the ratio of the number of characters of the candidate to the number of characters of the other candidate is less than a preset removal ratio, the candidate is Removing, for each possible combination of each category candidate stored in the extraction candidate storage means, calculating a feature vector representing the characteristics of the combination, and stored in the extraction model storage means Obtaining a combination having a maximum score calculated using the extracted model and the feature vector; The combination of the scores obtained in the step of calculating the score by reading the combination of correct answers from the correct answer storage means storing the score and obtaining the combination with the maximum score for the score of the correct combination is within a predetermined range. In this case, the method includes a step of updating the extraction model stored in the extraction model storage unit, and stored in the prediction target extraction candidate storage unit that extracts and stores candidates for each category from the prediction target document information. For all possible combinations of candidates for each category, a feature vector representing the characteristics of the combination is calculated, and a score calculated using the extracted model stored in the extracted model storage means and the feature vector is It has the step which acquires the combination which becomes the largest, It is characterized by the above-mentioned.

  In the information extraction method, the step of updating the extraction model reads the error cost from cost storage means storing error costs when the candidate is the correct partial character string and when it is not the partial character string, and A cost according to the proportion of the candidates corresponding to the correct answer is obtained, and the obtained cost is set as the predetermined range.

  An information extraction program according to a third aspect of the present invention is characterized in that a computer is operated as each means of the information extraction apparatus.

  ADVANTAGE OF THE INVENTION According to this invention, when extracting the information of the category which is relevant from a document, an appropriate candidate is extracted and the increase in learning cost can be prevented.

It is a functional block diagram which shows the structure of the event information extraction apparatus in this Embodiment. It is a figure which shows the example of the data stored in the document database. It is a figure which shows the example of the data stored in the extraction candidate database. It is a figure which shows the example of the data stored in the correct database. It is a figure which shows the example of the data stored in the cost database. It is a flowchart which shows the flow of a process of an event name candidate extraction function. It is a flowchart which shows the flow of a process of an event extraction model learning function. It is a figure which shows the example of the event extraction model stored in the event extraction model database. It is a figure which shows the example of the data stored in the prediction object document database. It is a figure which shows the example of the data stored in the prediction object extraction candidate database. It is a flowchart which shows the flow of a process of an event extraction function. It is a figure which shows the example of the event information stored in the event database.

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

  FIG. 1 is a functional block diagram showing the configuration of the event information extraction apparatus in the present embodiment. 1 includes a document DB 10, an event name candidate extraction function 15 (corresponding to candidate removal means), an extraction candidate DB 20 (corresponding to extraction candidate storage means), a correct answer DB 30 (corresponding to correct answer storage means), a cost. DB 40 (corresponding to cost storage means), event extraction model learning function 50 (corresponding to feature vector calculation means, combination acquisition means and extraction model updating means), event extraction model DB 60 (corresponding to extraction model storage means), prediction target document DB 70 A prediction target extraction candidate DB 80 (corresponding to the prediction target extraction candidate storage means), an event extraction function 90 (corresponding to the information extraction means), and an event DB 100. Each unit included in the event information extraction device may be configured by a computer including an arithmetic processing device, a storage device, and the like, and the processing of each unit may be executed by a program. This program is stored in a storage device included in the event information extraction device, and can be recorded on a recording medium such as a magnetic disk, an optical disk, or a semiconductor memory, or provided through a network.

  First, the information stored in the document DB 10, the extraction candidate DB 20, the correct answer DB 30, and the cost DB 40 used in the process of generating the event extraction model will be described.

  The document DB 10 assigns and stores a document ID to text information (text data) from which event information is extracted. FIG. 2 shows an example of data stored in the document DB 10.

  The extraction candidate DB 20 stores event name candidates, location candidates, and date / time candidates extracted from each of the text information stored in the document DB 10 for each document ID. FIG. 3 shows an example of data stored in the extraction candidate DB 20. In the example illustrated in FIG. 3, “event” and “Otaru Long Christmas 2012 Final” are stored as event name candidates extracted from the body information with the document ID of 1. Although not shown in FIG. 3, appearance position information in the text information is given to event name candidates, place candidates, and date / time candidates. The event name candidate is extracted from the text information stored in the document DB 10 by the event name candidate extraction function 15. By using Non-Patent Document 2 for extracting location candidates and Non-Patent Document 3 for extracting date / time candidates, it is possible to generate data to be stored in the extraction candidate DB 20 from text information stored in the document DB 10. If the event name, location, and date / time can be extracted, event recommendation based on the location / date / time is possible. In this embodiment, these three pieces of information are used as a unit of event information. The event name, location, date and time are called event information categories. In the present embodiment, the event name, location, and date / time are used as the event information category, but other information (for example, fee, host organization, etc.) can be realized in the same framework as in the present embodiment.

  The correct answer DB 30 stores the correct answer of each category, that is, the correct event name, the correct answer place, and the correct date and time for each of the text information stored in the document DB 10. FIG. 4 shows an example of data stored in the correct answer DB 30. These correct answers are prepared in advance by hand. In addition, like the extraction candidate DB 20, appearance position information in the text information is given.

  The cost DB 40 stores the cost for the error type. FIG. 5 shows an example of data stored in the cost DB 40. In FIG. 5, substring is the error cost when the correct event name is a substring, and others indicates the error cost for other errors. The value of substring is assumed to be set smaller than the value of others. These costs are preliminarily created manually.

  Next, the process flow of the event name candidate extraction function 15 will be described.

  FIG. 6 is a flowchart showing a process flow of the event name candidate extraction function 15.

  First, an unprocessed record is selected from the document DB 10 (step S11).

  Event name candidates are extracted from the text data of the selected record (step S12). To extract event name candidates, text data is divided into words and part-of-speech information is added to each word (for example, Takeshi Fuchi, “Japanese Morphological Analyzer using Word Co-occurence-JTAG”, COLING-ACL, pp.409- 413, 1998), based on the obtained part-of-speech information, consecutive nouns and word strings connected by “no” are all set as event name candidate sets S. For example, when word division is performed on consecutive nouns such as “20th”, “Yokosuka”, and “Festival”, the following six event names are extracted.

・ 20th ・ 20th Yokosuka ・ 20th Yokosuka Festival ・ Yokosuka ・ Yokosuka Festival ・ Festival

  Then, from the obtained event name candidate set S, a part of candidates that are partial character strings is removed (step S13). As the number of event name candidates increases, the number of possible candidate combinations in all categories increases, which increases the cost of model generation in the event extraction model learning function 50. Therefore, in this embodiment, a part of event name candidates is removed. As a removal method, if the event name candidate is a partial character string of another event name candidate using a removal ratio 0 <λ ≦ 1 set in advance by hand, the number of characters of another event name candidate When the ratio of the number of characters of the partial character string to the number is less than the removal ratio λ, the partial character string is removed. For example, when λ = 1, all event name candidates other than the “20th Yokosuka Festival” are removed in the above example.

  The event name candidate obtained in step S13 is output to the corresponding record in the extraction candidate DB 20 (step S14).

  If there is an unprocessed record in the document DB 10 (Yes in step S15), the process returns to step S11 to acquire the next record, and if there is no unprocessed record (No in step S15), the process ends. .

  Next, the process flow of the event extraction model learning function 50 will be described.

  FIG. 7 is a flowchart showing a processing flow of the event extraction model learning function 50.

First, the weight vector w is initialized as w = (0, 0, 0,..., 0) ^T, and the repetition counter t is initialized as t ← 1 (step S21). The weight vector w is an event extraction model stored in the event extraction model DB 60. The number of dimensions of the weight vector w is the same as the number of feature vectors described later.

  Subsequently, one record is selected at random from the correct answer DB 30 (step S22). The document ID of the record selected here is d.

  Subsequently, a record with the document ID d is selected from the extraction candidate DB 20, and a set of possible candidate combinations in all categories (hereinafter referred to as "category combination set") is created and included in the category combination set. Feature vectors are created for all combinations (step S23). The category combination set created from the record whose document ID is 2 in the extraction candidate DB 20 shown in FIG. 3 is as follows.

The 20th Yokosuka Festival-Yokosuka City, Kanagawa Prefecture-October 20, 2012 The 20th Yokosuka Festival-Yokosuka City, Kanagawa Prefecture-December 20, 2012 ・ ・ ・
Festival-Tokyo-January 1, 2013

  Thus, the category combination set in the present embodiment is a set of all possible candidate combinations for event name candidates, place candidates, and date / time candidates. In the example of document ID 2 in FIG. 3, there are two event name candidates, two place candidates, and three date / time candidates, so there are 2 · 2 · 3 = 12 combinations. Here, the combination having the event name, location, and date / time stored in the selected record of the correct answer DB 30 is treated as a correct answer, and other combinations are regarded as incorrect combinations, and information on which category is incorrect is held simultaneously. It shall be. For example, if the location and date / time are different from the correct answer, it is determined that the location / date / time is incorrect. Thus, an error combination has one or more incorrect categories.

  In step S23, a feature vector Φ (y, x) representing the feature of each category combination is further created based on the created category combination set and the text information acquired from the document DB 10. Here, x is a vector representation of text information corresponding to the document ID, and y is an element (category combination) of the category combination set. Φ (y, x) is an M-dimensional vector, and is composed of an output of a feature function φ (y, x) for capturing the likelihood of M pieces of event information having y and x as inputs. As an example of the feature function φ (y, x), for example, in order to capture the feature that “three candidates of y appear in close positions in the document”, 1 when three expressions appear within 50 characters, Otherwise, a feature function that outputs 0 is given. Another example is a feature function that outputs 1 when a character string included in an event name also appears elsewhere in the text, and outputs 0 otherwise. As a basic feature function based on other character strings, for example, the method of Non-Patent Document 4 can be used.

Subsequently, the cost for an error in the event name is calculated using the cost stored in the cost DB 40 (step S24). Specifically, when the event name of the category combination set is a partial character string of the correct event name, the value of substring in the cost DB 40 is set to cost _substring, and the cost is calculated using the following equation (1).

Here, length _substring represents the character string length of the partial character string, and length _{correct_string} represents the character string length of the correct event name. Since the cost value increases as the partial character string length becomes shorter, the effect of a penalty term that does not generate a model for selecting such event name candidates is achieved.

  If the event name of the category combination set is not a partial character string of the correct event name, the value of others in the cost DB 40 is used as the cost. Since the value of others is larger than the value of substring, the error that is a substring is selected compared to the error that the event name of the category combination set is not a substring of the correct event name. This produces the effect of generating a model that selects a character string closer to the correct event name even in the case of an error.

Subsequently, the category combination that provides the maximum score with the current weight vector w is obtained (step S25). The category combination that gives the maximum score is calculated by the following equation (2).

Here, Y _t is the category combination sets in the document selected in the t-th iteration, the y _t category combinations of answer, is x _t is the body information of the document. However, it is not necessary to add cost.

Subsequently, the loss of the category combination of the maximum score obtained in step S25 is calculated. If the loss is greater than 0, the weight vector w is updated (step S26). The loss l _t in the t-th iteration is calculated by the following equation (3).

When the loss l _t > 0, the weight vector w is updated according to the loss l _t . For example, the method of Non-Patent Document 5 can be used to update the weight vector.

  Then, the repeat counter t is incremented (step S27), and if it is less than or equal to a predetermined repeat count T (Yes in step S28), the process returns to step S22, and if the repeat counter t exceeds the repeat count T (No in step S28). ), And outputs the weight vector w to the event extraction model DB 60 (step S29).

  Next, the event extraction function 90 will be described. The event extraction function 90 extracts event information from information stored in the prediction target document DB 70 and the prediction target extraction candidate DB 80 using the event extraction model stored in the event extraction model DB 60.

The event extraction model DB 60 stores the event extraction model obtained by the event extraction model learning function 50. The event extraction model is composed of weight vectors w = (w ₁ , w ₂ ,..., W _M ) ^T for M-dimensional features. FIG. 8 shows an example of an event extraction model stored in the event extraction model DB 60.

  Similar to the document DB 10, the prediction target document DB 70 assigns and stores a document ID to the text information from which event information is extracted. FIG. 9 shows an example of data stored in the prediction target document DB 70.

  Like the extraction candidate DB 20, the prediction target extraction candidate DB 80 stores event name candidates, place candidates, and date / time candidates extracted from the text information stored in the prediction target document DB 70. FIG. 10 shows an example of data stored in the prediction target extraction candidate DB 80.

  Next, the process flow of the event extraction function 90 will be described.

  FIG. 11 is a flowchart showing a process flow of the event extraction function 90.

  First, an unprocessed record is selected from the prediction target document DB 70 (step S31). It is assumed that the document ID of the record selected here is d ′.

  A record with a document ID of d 'is selected from the prediction target extraction candidate DB 80 to create a category combination set, and feature vectors are created for all category combinations included in the category combination set (step S32). A feature vector Φ (y, x) is created by the same processing as step S13 by the event extraction model learning function 50.

Subsequently, the category combination that gives the maximum score is acquired using the event extraction model stored in the event extraction model DB 60 (step S33). Specifically, as shown in the following equation (4), the inner product of the weight vector w stored in the event extraction model DB 60 and the feature vector Φ (y, x) created in step S32 is calculated, and the maximum Get the category combination that gives the score.

Here, Y _test is a set of category combinations in the input document, and x is text information of the input document.

  Each category of the category combination acquired in step S33 is output to the event DB 100 (step S34).

  If there is an unprocessed record in the prediction target document DB 70 (Yes in step S35), the process returns to step S31 to acquire the next record. If there is no unprocessed record (No in step S35), the process is performed. finish.

  FIG. 12 shows an example of event information stored in the event DB 100. The event DB 100 stores event information extracted for each document ID.

  As described above, according to the present embodiment, when the event name candidate extraction function 15 is a partial character string of another event name candidate, the number of characters of another event name candidate is determined. When the ratio of the number of characters in the partial character string is less than the removal ratio λ, by extracting the partial character string, an appropriate event name candidate is extracted, and the number of event name candidates is reduced to prevent an increase in learning cost. Can do.

  According to the present embodiment, by obtaining an error cost according to the rate of match of event name candidates to correct event names, a candidate for a partial character string of correct event names can be compared with a shorter partial character string. Thus, the error cost can be increased and the error cost can be set lower than an error that is not a partial character string.

10 ... Document DB
15 ... Event name candidate extraction function 20 ... Extraction candidate DB
30 ... Correct DB
40 ... Cost DB
50 ... Event extraction model learning function 60 ... Event extraction model DB
70 ... prediction target document DB
80 ... prediction target extraction candidate DB
90 ... Event extraction function 100 ... Event DB

Claims (5)

Extraction candidate storage means for extracting and storing candidates for each relevant category from document information;
Among the candidates extracted from the document information, when the candidate is a partial character string of another candidate of the same category, the ratio of the number of characters of the candidate to the number of characters of the other candidate is a preset removal ratio. Candidate removal means for removing the candidate if not,
Correct answer storage means storing correct answers for each category;
Extraction model storage means for storing an extraction model for extracting relevant category information;
For all possible combinations of the extraction candidate storage means for each stored category candidates, the feature vector calculating means for calculating a feature vector indicating the feature of the combination,
A combination acquisition unit that acquires a combination with a maximum score calculated using the extraction model stored in the extraction model storage unit and the feature vector;
When the combination of the correct answers acquired by the combination acquisition means with respect to the correct combination score is within a predetermined range by reading the correct combination from the correct storage means and calculating a score, the extracted model storage means An extraction model updating means for updating the extraction model stored in
Prediction target extraction candidate storage means for extracting and storing candidates for each category from the prediction target document information;
For all possible combinations of the prediction target extraction candidate storage unit the category of each stored in the candidate, a feature vector indicating the feature of the combination is calculated, and extracted model stored in the extraction model storage means And an information extraction unit that obtains a combination having a maximum score calculated using the feature vector. Cost storage means for storing an error cost when the candidate is the correct partial character string and when it is not a partial character string,
The extraction model update unit reads the error cost from the cost storage unit, obtains a cost according to a ratio of the candidate match to the correct answer, and sets the obtained cost as the predetermined range. Item 1. The information extraction device according to Item 1. An information extraction method executed by a computer,
Among the candidates stored in the extraction candidate storage means that extracts and stores candidates for each related category from the document information, when the candidate is a partial character string of another candidate of the same category, Removing the candidate when the ratio of the number of characters of the candidate to the number of characters of the other candidate is less than a preset removal ratio;
Calculating, for all possible combinations of candidates for each category stored in the extraction candidate storage means, a feature vector representing the characteristics of the combination;
Obtaining a combination that maximizes the score calculated using the extracted model and the feature vector stored in the extracted model storage means;
The loss of the combination score acquired in the step of calculating the score by reading the combination of the correct answers from the correct storage means storing the correct answers of each category, and acquiring the combination that maximizes the score with respect to the score of the correct combination Is within a predetermined range, updating the extraction model stored in the extraction model storage means,
For each possible combination of candidates for each category stored in the prediction target extraction candidate storage means that extracts and stores candidates for each category from the prediction target document information, a feature vector representing the characteristics of the combination is calculated. And obtaining a combination that maximizes the score calculated using the extraction model stored in the extraction model storage means and the feature vector.   The step of updating the extraction model is performed by reading the error cost from cost storage means storing error costs when the candidate is the correct partial character string and when the candidate is not a partial character string, and matching the candidate with the correct answer The information extraction method according to claim 3, wherein a cost corresponding to the ratio is obtained, and the obtained cost is set to the predetermined range.   An information extraction program for operating a computer as each means of the information extraction apparatus according to claim 1.

Images