JP2009217642A - Reputation information extracting device, dictionary constructing method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically register a new expression in a dictionary in extraction processing of reputation information. <P>SOLUTION: This reputation information extracting device having a function for extracting reputation information being a set of an evaluation expression and an attribute expression from text by referring to an evaluation dictionary and an attribute information includess a means for determining that one clause is an additional candidate clause to the attribute dictionary when it is determined that the other clause between two clause constituting a clause pair extracted from a sentence includes an evaluation expression in the evaluation dictionary and that the one clause includes no expression in any dictionary, and for determining that the one clause is an additional candidate clause to the evaluation dictionary when it is determined that the other clause includes an attribute expression in the attribute dictionary and that the one clause does include no expression in any dictionary, and a means for extracting an evaluation expression or an attribute expression from the additional candidate clause to the evaluation dictionary or the additional candidate clause to the attribute dictionary and adding the extracted expression as a new expression to the evaluation dictionary or the attribute dictionary. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a technique for extracting reputation information from a group of sentences stored on a review site or the like provided on a communication network such as the Internet.

  In recent years, CGM (Customer Generated Media) content that uses individuals as an information source has been increasing, and among them, websites (review sites, etc.) where evaluations of services, products, etc. are described by general consumers It has become particularly popular.

  Such websites contain text that includes a reputation such as “meat is juicy” for a particular object (eg, a particular store). If you can get a pair of evaluation expression “juicy” and attribute expression “meat” corresponding to “what” for the evaluation expression as reputation information from sentences including such reputation, use the service Useful when purchasing products.

As a conventional technique for extracting reputation information composed of a set of evaluation expression and attribute expression as described above from a sentence group related to a certain object, a list of attribute expressions (attribute dictionary) and a list of evaluation expressions (evaluation dictionary) are preliminarily stored. There is a technique for extracting reputation information composed of a set of an evaluation expression and an attribute expression from a sentence group by pattern matching by creating an attribute dictionary and an evaluation dictionary. In addition, there exists patent document 1 as a prior art document relevant to this application.
JP 2007-115118 A

  However, if the attribute dictionary and the evaluation dictionary are manually created, a large cost is required, and it is not easy to provide reputation information extraction as an actual service. From this point of view, it would be very convenient to extract reputation information from a group of sentences on a review site where additional sentences are accumulated daily and to automatically create an attribute dictionary and an evaluation dictionary.

  The present invention has been made in view of the above points, extracts reputation information from a sentence group using an attribute dictionary and an evaluation dictionary, extracts attribute expressions and evaluation expressions to be newly registered, and extracts them. An object of the present invention is to provide a technique that enables automatic registration in an attribute dictionary and an evaluation dictionary.

  In order to solve the above-described problems, the present invention includes an evaluation dictionary storage unit that stores an evaluation dictionary and an attribute dictionary storage unit that stores an attribute dictionary, and stores text by referring to the evaluation dictionary and the attribute dictionary. A reputation information extracting device having a function of extracting reputation information which is a set of an evaluation expression and an attribute expression from text stored in the means, wherein the text stored in the text storage means is read out and included in the text A phrase pair extraction unit that performs dependency analysis of a sentence and extracts a pair of phrases having a dependency relation from the sentence, and compares each phrase of the phrase pair with expressions included in the evaluation dictionary and the attribute dictionary The comparison means for performing the above and the comparison means, one of the two clauses constituting the clause pair includes an evaluation expression in the evaluation dictionary, and the other clause is If it is determined that the expression in the dictionary of the deviation is not included, the other phrase is determined to be an attribute dictionary addition candidate phrase including an attribute expression to be added to the attribute dictionary, and the comparison means An evaluation expression to be added to the evaluation dictionary when it is determined that the other phrase includes an attribute expression in the attribute dictionary and the other phrase does not include an expression in any dictionary. The dictionary addition candidate clause determining means for determining that the evaluation dictionary addition candidate clause includes and the phrase pair including the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause satisfy the predetermined condition based on the part of speech, A new expression adding means for extracting an evaluation expression or attribute expression from the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause and adding it as a new expression to the evaluation dictionary or the attribute dictionary; It was example configured as reputation information extracting apparatus according to claim.

  The new expression adding means determines whether the part of speech of the last word of the first phrase of two phrases in the phrase pair including the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause satisfies a predetermined condition A phrase pair filtering unit for determining whether or not the phrase pair filtering unit adds the new expression when it is determined that at least a part of speech of the last word of the first phrase satisfies a predetermined condition. The means may add an evaluation expression included in the evaluation dictionary addition candidate clause or an attribute expression included in the attribute dictionary addition candidate clause to the evaluation dictionary or the attribute dictionary as the new expression.

  In the case where the first phrase is a phrase including the attribute dictionary addition candidate phrase or the attribute expression in the attribute dictionary, the predetermined condition is that at least the part of speech of the last word is a case particle or a coordinator. In the case where the first phrase is a phrase including the evaluation dictionary addition candidate clause or the evaluation expression in the evaluation dictionary, the predetermined condition is that the part of speech of the last word has a utilization form. It may be a part of speech, and the utilization form may be a basic form or a body connection.

  In addition, when the phrase pair satisfies the condition in the phrase pair filtering means, the new expression adding means is the part of speech of the last word of the expression extracted from the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause Whether or not a predetermined condition is satisfied, and when the condition is satisfied, the expression may be added to the evaluation dictionary or the attribute dictionary. Here, when the expression is an evaluation expression, the predetermined condition is that the last word of the evaluation expression is a noun that is a general noun, an adjective, or an adjective verb stem, and the expression is an attribute expression. In this case, the predetermined condition may be that the last word of the attribute expression is a general noun or proper noun.

  In the reputation information extracting device, one of two clauses constituting the clause pair includes an evaluation expression in the evaluation dictionary, and the other clause is an attribute in the attribute dictionary. When it is determined that an expression is included, an evaluation expression is extracted from the one phrase, an attribute expression is extracted from the other phrase, and a set of the evaluation expression and the attribute expression is stored as reputation information in the reputation information storage unit. Reputation information extraction means for storing may be further provided.

  The present invention can also be configured as a program that causes a computer to function as each processing means in the above reputation information extraction apparatus.

  According to the present invention, reputation information is extracted from a sentence group using an attribute dictionary and an evaluation dictionary, and an attribute expression and an evaluation expression to be newly registered are extracted, and these are automatically converted into the attribute dictionary and the evaluation dictionary. It becomes possible to register.

  Embodiments of the present invention will be described below with reference to the drawings. In the present embodiment, an example in which the present invention is applied to a group of sentences accumulated at a restaurant review site is described. However, the present invention is not limited to a group of sentences in a specific area (such as a restaurant). Needless to say, the present invention can be applied to a sentence group in any region.

(System configuration)
FIG. 1 shows a system configuration in the embodiment of the present invention. As shown in FIG. 1, the system according to the embodiment of the present invention extracts a reputation information by executing a process according to the present invention, a review site providing apparatus 1 that provides a restaurant review site to a user terminal 3. A reputation information extraction device 2 that constructs an attribute dictionary / evaluation dictionary and a user terminal 3 are connected to a communication network 4 such as the Internet.

  FIG. 2 shows a functional configuration diagram of the review site providing apparatus 1. As shown in FIG. 2, the review site providing apparatus 1 includes a user terminal interface unit 11, a review information accumulation / providing unit 12, a reputation information providing unit 13, a reputation information extracting device interface unit 14, a text storage unit 15, reputation information. A storage unit 16 is included.

  The user terminal interface unit 11 is a functional unit for communication with the user terminal 3, provides a screen for writing and searching for review text to the user terminal 3, and information transmitted from the user terminal 3. Receive.

  The review information accumulation / providing unit 12 stores the review text received from the user terminal 3 via the user terminal interface unit 11 in the text storage unit 15 or reviews from the text storage unit 15 in response to a request from the user terminal 3. It is a functional unit for reading a sentence and transmitting it to the user terminal 3 through the user terminal interface unit 11 together with the reputation information received from the reputation information providing unit 13.

  The reputation information providing unit 13 is a functional unit that acquires reputation information from the reputation information storage unit 16 in response to a request from the user terminal 3 and transmits it to the review information storage and provision unit 12.

  The reputation information extracting device interface unit 14 is a functional unit for communicating with the reputation information extracting device 2, reads text from the text storage unit 15 in response to a request from the reputation information extracting device 2, and converts it into reputation information. It is a functional unit that transmits reputation information to the extraction device 2 or receives reputation information from the reputation information extraction device 2 and stores it in the reputation information storage unit 16.

  FIG. 3 shows an example of the structure of data stored in the text storage unit 15. As shown in FIG. 3, the text storage unit 15 stores a group of sentences written by the user as a review in association with the name of the store that is the subject of reputation. In FIG. 3, each sentence increases every time a user writes a review. This text may be categorized according to the type of restaurant (Chinese, French, etc.).

  FIG. 4 shows an example of the structure of data stored in the reputation information storage unit 16. As shown in FIG. 4, the reputation information storage unit 16 stores reputation information made up of a set of attribute expressions and evaluation expressions in association with the store name. This reputation information is created by the reputation information extraction device 2 based on the text and is stored by being received from the reputation information extraction device 2.

  When the user terminal 3 accesses the review site providing apparatus 1 and requests, for example, the review information of the store A, the review information accumulation / providing unit 12 acquires a review text group corresponding to the store A from the text storage unit 15. At the same time, the reputation information acquisition unit 13 acquires the reputation information corresponding to the store A from the reputation information storage unit 16 and passes it to the review information storage and provision unit 12. Then, the review information accumulation / providing unit 12 displays a screen as shown in FIG. 5 on the user terminal 3, for example. In addition to the display format shown in FIG. 5, reputation information may be displayed in a tag cloud format. In this case, for example, an evaluation expression having a high appearance frequency is displayed in large characters, and an attribute expression corresponding to the evaluation expression is displayed below the large expression.

  FIG. 6 shows a functional configuration diagram of the reputation information extraction device 2. As shown in FIG. 6, the reputation information extraction device 2 includes an anti-review site providing device interface unit 21, a text acquisition unit 22, a preprocessing unit 23, a dependency analysis unit 24, a phrase pair acquisition unit 25, and a phrase pair filtering unit 26. An expression extraction unit 27, an expression filtering unit 28, a data storage unit 29, a dictionary storage unit 30, a reputation information storage unit 31, and a reputation information providing unit 32.

  The review site providing apparatus interface unit 21 is a functional unit for communicating with the review site providing apparatus 1 via the communication network 4. The text acquisition unit 22 acquires text (hereinafter referred to as extraction target text) that is a set of sentences from which the evaluation expression and attribute expression are extracted from the review site providing apparatus 1, and the acquired text is stored in the data storage unit 29. It is a functional part for storing in (which is also a text storage part). Note that the text acquisition unit 22 may designate a certain area (Chinese etc.) and acquire the text in that area as a processing target.

  The pre-processing unit 23 is a functional unit that divides the acquired text into a plurality of sentences and performs a shaping process on the divided text. The dependency analysis unit 24 is a functional unit that performs dependency analysis on the sentence that has been processed by the preprocessing unit 23. The phrase pair acquisition unit 25 (including the dictionary addition candidate phrase determination unit), based on the result obtained by the dependency analysis, a phrase including the expression in the evaluation dictionary or the attribute dictionary, and a phrase having a dependency relationship with the phrase Is a functional unit that acquires as a pair.

  The phrase pair filtering unit 26 performs morphological analysis on the phrases in each phrase pair obtained by the phrase pair acquisition unit 25, and leaves only the phrase pairs that match a predetermined part-of-speech condition, which are the targets of subsequent processing. It is a functional part.

  The expression extraction unit 27 is a functional unit that extracts an evaluation expression from a phrase including an evaluation expression and extracts an attribute expression from a phrase corresponding to the attribute expression for a phrase pair that matches the condition in the phrase pair filtering unit 26.

  The expression filtering unit 28 performs a morphological analysis on each of the evaluation expression and attribute expression extracted by the expression extraction unit 27, and leaves a pair of evaluation expression and attribute expression whose part of speech of the last word matches a predetermined condition. Part.

  The data storage unit 29 is a functional unit that stores text received from the review site providing apparatus 1 and stores data of processing results in each processing functional unit. Each processing function unit reads and processes the previous stage processing result stored in the data storage unit 29, and stores the processing result in the data storage unit 29.

  The dictionary storage unit 30 is a functional unit that stores an attribute dictionary and an evaluation dictionary. More specifically, the dictionary storage unit 30 includes an attribute dictionary storage unit and an evaluation dictionary storage unit. The reputation information storage unit 31 is a functional unit for storing attribute expressions and evaluation expressions extracted as reputation information. The reputation information providing unit 32 is a functional unit for reading out reputation information from the reputation information storage unit 31 and transmitting it to the review site providing apparatus 1.

  In the present specification and claims, the attribute dictionary and the evaluation dictionary are a list of attribute expressions and a list of evaluation expressions. In addition, when “dictionary” is described, it is not particularly limited as to whether it is an attribute dictionary or an evaluation dictionary, and it is assumed to be either of these. In addition, when “expression” is described, it is not particularly limited whether it is an attribute expression or an evaluation expression, and it is any one of these. Further, in the present specification and claims, dictionary, expression, phrase, word, candidate, etc. mean dictionary data, expression data, phrase data, word data, candidate data, etc., and each is processed by a computer. It is assumed that the data is in a possible format.

  The reputation information extraction device 2 is realized by mounting a program for executing processing described in the present embodiment on a computer having a CPU, a storage device, a communication function, and the like. Each processing function unit in the reputation information extraction device 2 is a function unit realized by cooperation of hardware resources (CPU, storage device, etc.) and a program in the computer. The program may be stored in a recording medium such as a memory and then installed in a computer from there, or may be downloaded from a server on a network.

  Next, the outline of the process of the reputation information extracting device 2 will be described with reference to FIG. First, the reputation information extraction device 2 acquires the text from which the evaluation expression and attribute expression are extracted from the review site providing device 1 (step 1). Then, the reputation information extraction device 2 refers to the pre-created evaluation dictionary including a small number of evaluation expressions and the pre-created attribute dictionary including a small number of attribute expressions according to the present invention with respect to the text. By performing processing, reputation information (a combination of attribute expression and evaluation expression) is extracted, a new expression to be added to the dictionary (referred to as a new expression) is extracted, and the new expression is added to the dictionary (step 2). 3). That is, the new expression is added to the dictionary storage unit 30 as dictionary data. In the example shown in FIG. 7, “price” and “fish” are newly added to the attribute dictionary, and “cheap” and “fresh” are added to the evaluation dictionary. In the above operation, reputation information and a new expression to be added to the dictionary are extracted at the same time. However, only the new expression to be added to the dictionary may be extracted for the purpose of automatic construction of the dictionary only.

(Operation of reputation information extraction device)
Next, the operation of the reputation information extracting apparatus 2 will be described in detail with reference to the flowcharts of FIGS. 8 to 10 and the specific examples shown in FIGS.

  As a premise of the following processing, it is assumed that “good” is already stored as an evaluation expression in the evaluation dictionary storage unit in the dictionary storage unit 30 and “soup” is already stored as an attribute expression in the attribute dictionary storage unit. . These are manually stored. Note that the evaluation dictionary and the attribute dictionary are expanded by executing the processing flow according to the present embodiment a plurality of times.

  In the flowchart shown in FIG. 8, first, the text acquisition unit 22 in the reputation information extraction device 2 acquires the text to be extracted (expression target text) from the review site providing device 1 and stores it in the data storage unit 29. (Step 11). As described above, the subsequent processing in each processing function unit is performed by reading processing result data from the data storage unit 29 and storing the processing result data. The extraction target text may be based on text corresponding to a specific target (store), or may be based on a certain area (Chinese). Moreover, it is good also as a text corresponding to a wider range (the whole restaurant). However, the data storage unit 29 stores a sentence included in the text for each store and each area.

  An example of the extraction target text is shown in FIG. A specific example will be described below assuming that processing is performed on the extraction target text shown in FIG.

  Subsequently, the preprocessing unit 23 divides the extraction target text read from the data storage unit 29 into sentences (step 12). In the example of the extraction target text shown in FIG. 11A, it is divided into two sentences as shown in FIG. 11B, and the following processing is performed for each sentence. In the following processing, the first sentence is targeted.

  Subsequently, the preprocessing unit 23 performs a shaping process on the sentence to be processed (step 13). Specifically, for example, processing such as removal of an Html tag, removal of a Wiki grammar tag, or conversion of half-width characters to full-width characters is performed. The sentence after the sentence shaping process is shown in FIG.

  Next, the dependency analysis unit 24 performs dependency analysis on the sentence that has undergone the shaping process (step 14). In the present embodiment, a program called cabocha is used as a program for performing dependency analysis processing in the dependency analysis unit 24, but the present invention is not limited to specific dependency analysis processing. A general dependency analysis process may be performed.

  Subsequently, the phrase pair acquisition unit 25 extracts phrase pairs having a dependency relationship based on the dependency analysis result of the sentence in step 14 (step 15). That is, in step 14, data indicating which clause and which clause are in a dependency relationship is acquired as an analysis result, and using the result, a pair of clauses in a dependency relationship from the processing target sentence is used as data. Extract. An example of a phrase pair acquired by the phrase pair acquisition unit 25 is shown in FIG. 12 (a), “this-store”, “store-“ ikaruga ”, etc.” are phrase pairs having a dependency relationship, and in step 15, such phrase pairs are extracted one by one. Then, the following processing is performed for each phrase pair.

  Next, the document pair acquisition unit 25 compares the words in each phrase included in the phrase pair that is the processing target with the expressions included in the attribute dictionary and the evaluation dictionary (step 16). It is determined whether or not the phrase includes an expression included in the attribute dictionary or the evaluation dictionary (step 17).

  For example, in the example of each phrase pair shown in FIG. 12A, as shown in FIG. 12B, the words enclosed in parentheses are included in the evaluation expression or attribute dictionary included in the evaluation dictionary. A word that matches the attribute expression. In addition, the parentheses in the figure are described for easy understanding of the matched word, and are not included in the clause. The same applies hereinafter. Also, if the word corresponding to the expression in the dictionary is in the utilization form in the clause to be processed, it is converted to the same form (basic form) as in the dictionary, and the converted word is the expression in the dictionary. If it matches, the word in the phrase and the expression in the dictionary match.

  Whether each phrase in the phrase pair includes an expression included in the dictionary can be divided into the following cases.

  (A) None of the clauses in the clause pair includes an expression included in the dictionary.

  (B) Both clauses in the clause pair include expressions included in only one of the attribute dictionary and the evaluation dictionary.

  (C) One clause of the phrase pair includes an evaluation expression in the evaluation dictionary, and the other clause of the phrase pair includes an attribute expression in the attribute dictionary.

  (D) One clause of the phrase pair includes an evaluation expression in the evaluation dictionary or an attribute expression in the attribute dictionary, but the other phrase in the phrase pair does not include any dictionary expression.

  In step 17, the phrase pair acquisition unit 25 determines which of (A) to (D) the target phrase pair corresponds to. If (A) or (B), the phrase pair acquisition unit 25 returns to step 15 to Process phrase pairs. The phrase pair acquisition unit 25 proceeds to the process of step 19 when the determination result in step 17 is (C), and proceeds to the process of step 18 when the determination result is (D).

  In step 18 in the case where the determination result is (D), the document pair acquisition unit 25 adds the phrase on the side that does not include the expression in the dictionary, out of the two phrases in the phrase pair, to the dictionary. It is regarded as a dictionary addition candidate clause including a candidate for a new expression to be added, and a flag (dictionary addition candidate flag) indicating this is given to the clause. More specifically, when one of the two clauses in the clause pair includes an evaluation expression in the evaluation dictionary, the document pair acquisition unit 25 includes a candidate for a new attribute expression in the other clause. If a flag indicating that it is an attribute dictionary addition candidate clause is given to the clause, and one of the two clauses in the clause pair includes the attribute expression in the attribute dictionary, the other clause Is included as a candidate for evaluation expression, and a flag indicating that it is an evaluation dictionary addition candidate clause is given to the clause. Then, the process proceeds to Step 19.

  FIG. 12C shows the result after the processing of Step 17 and Step 18 is performed on each sentence shown in FIG. Flags are shown using small parentheses such as "(attribute)". In the flow of the present embodiment, processing is performed for each clause pair. Here, in order to make the processing easy to understand, processing results for all clause pairs are shown.

  As shown in FIG. 12C, for example, the phrase “soup of“ ikaruga ”is related to the phrase pair, and the phrase“ ikaruga ”does not include any dictionary expression, but the phrase“ soup ”is Since the attribute expression is included in the attribute dictionary, a flag indicating that it is an evaluation dictionary addition candidate clause is assigned to “of ikaruga”. The determination in 17 is a pair corresponding to (C).

  Subsequently, the document pair acquisition unit 25 checks the phrase engagement method for the phrase pair that has passed through step 17 or step 18 (step 19). In this embodiment, the subsequent processing is performed on the phrase pairs corresponding to the above (C) and (D). However, for example, when only the dictionary construction is to be performed, the above (D) The subsequent processing may be performed only for the phrase pair to which the flag corresponding to is attached.

  Specifically, in the check on how to handle clauses, the document pair acquisition unit 25 determines whether the dependency distance is not too long, whether the dependency format is parallel or not equivalent, and the words constituting the clause of the phrase pair Check if there are too many numbers. To determine whether the dependency distance is too long, for example, in the original sentence, calculate the number of words by morphologically analyzing the phrase included between one phrase and the other phrase in the phrase pair. Can be determined by whether or not is greater than or equal to a predetermined number. Instead of using the number of words in this way, it may be determined whether the distance is too long by using the number of clauses included between one clause and the other.

  Whether the number of words constituting the phrase is too large can be determined by calculating the number of words in the phrase by morphological analysis of the phrase and determining whether the number of words is equal to or greater than a predetermined number. If the phrase pair to be processed does not satisfy the above check in step 19, the process proceeds to step 15 where the next phrase pair is processed.

  Next, the phrase pair filtering unit 26 checks the dependency format based on the part of speech for the phrase pair that has passed through step 19 (step 20 in FIG. 9). Specifically, the document pair filtering unit 26 performs a check according to the procedure shown in FIG. As shown in FIG. 10, first, the document pair filtering unit 26 refers to the contents of the first phrase and the next phrase in the phrase pair, and checks the order in which the phrases appear (step 201). If the first clause is a clause including an attribute expression and the next clause is a clause including an evaluation expression, the process of step 202 is performed, and the first clause is a phrase including an evaluation expression. If the phrase is a phrase including an attribute expression, the process of step 203 is performed. The phrase pair filtering unit 26 determines whether or not the part of speech of the last word of the first phrase of two phrases in the phrase pair including the evaluation dictionary addition candidate phrase or the attribute dictionary addition candidate phrase satisfies a predetermined condition. This corresponds to the phrase pair filtering means for determination.

  Whether or not the phrase includes an attribute expression is determined based on whether or not the phrase includes an attribute expression included in the attribute dictionary or whether the phrase is an attribute dictionary addition candidate phrase. Whether the phrase includes an evaluation expression is determined by whether the phrase includes an evaluation expression included in the evaluation dictionary or whether the phrase is an evaluation dictionary addition candidate phrase.

  In step 202 when the first clause is a clause including an attribute expression and the next clause is a clause including an evaluation expression, the document pair filtering unit 26 performs morphological analysis on the clause including the attribute expression, and Whether the last part of speech of the word is “particle-case particle” or “particle-coordinate particle” and it is one of “ga”, “ha”, “mo” If the determination result is OK, the process proceeds to the next process, but if it is NG, the process returns to step 15 to proceed to the next phrase pair process.

  In this embodiment, a program called mecab is used as a morphological analysis program, and the part-of-speech system used in this embodiment is based on the IPA part-of-speech system used in mecab. The usage form is also used by mecab. However, for example, “particle-case particle” merely means a case particle in a particle, that is, a case particle, and the type of part of speech described in the present embodiment is a general Japanese grammar. It is what I followed. The usage form also follows general Japanese grammar. The term “particle-case particle” means a case particle in which the part of speech as a particle is subdivided into a connected particle, a case particle, an auxiliary particle, and the like. The same applies hereinafter.

  In step 202, for example, when the phrase pair is “meat is juicy”, the part of speech of the word “ga” that appears at the end of the phrase “meat” including the attribute expression is a case particle. The determination at is OK. When the phrase pair is “Meat / Producing region”, “NO” in “Meat” does not match the condition of step 202, so the determination is NG.

  In step 203 when the first clause is a clause containing an evaluation expression and the next clause is a clause containing an attribute expression, the clause pair filtering unit 26 performs a morphological analysis on the clause containing the evaluation expression, It is determined whether or not the last word usage form is “basic form” or “diagram connection”. If the result of the determination is OK, the process proceeds to the next process. Proceed to processing the phrase pair. That is, here, it is determined whether or not the last word in the phrase is a part-of-speech word having an inflected form, and the inflected form is “basic form” or “partially connected”. Note that the “dialogue connection” is a utilization form that connects to the paraphrase. In steps 202 and 203, if the word to be checked is a symbol, the check is performed with the previous word.

  FIG. 13A shows the determination results for the example of each phrase pair shown in FIG. For example, in the phrase pair ““ ikaruga ”-soup is”, the first phrase contains an evaluation expression, so the determination in step 203 is performed, but there is no utilization form of the last word “no” in the phrase. The determination at 203 is NG. On the other hand, if the phrase pair is “Soup is good,” the attribute is included in the first phrase, so the determination in step 202 is made. And since the last word of “Soup” is “ha” and the part of speech is an auxiliary particle, the determination here is OK.

  In step 20 in FIG. 9, after the determination result of OK is obtained for the phrase pair to be processed, the expression extracting unit 27 includes each phrase (evaluation dictionary) including the evaluation expression or attribute expression in the phrase pair. Additional candidate clauses or attribute dictionary additional candidate clauses), the part of speech of the words in the clause is checked in order from the word at the end of the clause, and in this processing, the part of speech of the word to be checked is If it is a symbol, particle, auxiliary verb, or conjunction, the word is deleted and the part of speech confirmation process for the next word is performed. If the part of speech of the word to be confirmed is neither a symbol, a particle, an auxiliary verb, nor a conjunction, the part of speech confirmation process ends at the time the word appears, and the remaining part of the word is not evaluated Extracted as an expression (step 21). Of course, the part remaining after performing the above processing in the clause including the evaluation expression is the evaluation expression, and the part remaining after performing the above processing in the clause including the attribute expression is the attribute expression.

  In the processing of step 21 above, in the case of a phrase including an evaluation expression, the auxiliary verb “none” is not subject to deletion. Further, when the last word of the portion remaining as a result of the processing is a utilization form, the last word is converted into a basic form.

  In the phrase pair determined to be OK in step 20 as shown in FIG. 13A, in step 21, the evaluation expression and the attribute expression are extracted as shown in FIG. 13B. For example, in the phrase pair "Soup-well," the phrase containing the attribute expression "Soup" is deleted because "is" is a particle, "Soup" is a symbol, particle, auxiliary verb, conjunction As a result, “soup” is extracted as an attribute expression. "Well," "" and "" are deleted because they are symbols, and "Well" is converted to the basic form and extracted as an evaluation expression as "Good".

  Subsequently, the expression filtering unit 28 performs morphological analysis on the extracted expression and confirms the part of speech of the last word (step 22). More specifically, the expression filtering unit 28 determines that the last word is “noun-general” (general noun), “adjective”, “noun-adjective verb stem” (“ It is determined whether it is any of the nouns that can take the form of “~” like “luxury”.

  When the confirmation target is an attribute expression, the expression filtering unit 28 determines whether the last word is “noun-general” or “noun-proper noun”.

  If the determination for both the evaluation expression and the attribute expression is OK, the determination in step 22 is OK, and the process proceeds to the next process. If the determination for either or both of the evaluation expression and the attribute expression is NG, the process proceeds to step 15. Return and process the next phrase pair. FIG. 13C shows the result of the determination in step 22 for each pair of attribute expression and evaluation expression shown in FIG. For example, regarding the pair of the evaluation expression “good” and the attribute expression “soup”, “good” is an adjective and “soup” is a general noun, so the determination result of step 22 is OK to satisfy the determination condition of step 22. It becomes.

  If the determination result in step 22 is OK, the expression filtering unit 28 determines whether a dictionary addition candidate flag is assigned to any of the attribute expression and evaluation expression pairs to be processed (step 23). . In the example of FIG. 13C, since the dictionary addition candidate flag is assigned to the attribute expression “thought”, if “good” and “thought” are processing targets, the determination result in step 23 is Yes, the expression filtering unit 28 adds a new attribute expression “thought” to the attribute dictionary storage unit (step 24). The expression filtering unit 28 determines the part of speech of the last word of the expression extracted from the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause when the phrase pair satisfies the corresponding condition in the phrase pair filtering unit 26. It is determined whether or not the condition is satisfied, and when the condition is satisfied, processing for adding the expression to the evaluation dictionary or the attribute dictionary is performed.

  If no flag is assigned to any attribute expression / evaluation expression pair to be processed in step 23, the expression filtering unit 28 stores the pair in the reputation information storage unit 31 (step 25). . In the example of FIG. 13C, as shown in FIG. 13D, a pair of the evaluation expression “good” and the attribute expression “soup” and a pair of the evaluation expression “good” and the attribute expression “chicken soup” It is stored in the reputation information storage unit 31 as reputation information. When storing reputation information in the reputation information storage unit 31, for example, the reputation information is stored in association with the store name corresponding to the sentence from which the reputation information is extracted.

  In step 26, it is determined whether or not the phrase pair to be processed is the last phrase in the sentence to be processed. If not, the process proceeds to the next phrase pair. If it is the last phrase pair, it is determined whether or not the sentence to be processed is the last sentence in the text (step 27). If it is not the last sentence, the process moves to the next sentence, and the last sentence If it is a sentence, the process ends.

  Thereafter, for example, in response to a request from the review site providing apparatus 1, the reputation information providing unit 32 reads out the reputation information from the reputation information storage unit 31 and transmits it to the review site providing apparatus 1. The above reputation information extraction / dictionary construction process may be performed, for example, every certain period (every time text is increased to some extent at the review site), or whenever the user writes on the review site, the text is processed. It is good also as sending to the reputation information extraction apparatus 2 from the review site provision apparatus 1, and the reputation information extraction apparatus 2 performing a reputation information extraction / dictionary construction process each time. Further, the review site providing apparatus 1 may be provided with the function of the reputation information extracting apparatus 2 in the present embodiment.

  As described above, according to the reputation information extraction apparatus 2 in the present embodiment, the evaluation dictionary and the attribute dictionary used for reputation information extraction are automatically created simply by registering a small number of words in advance. Is possible.

  The present invention is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the claims.

It is a system configuration figure in an embodiment of the invention. It is a functional block diagram of a review site provision apparatus. It is a figure which shows the structural example of the data which a text storage part stores. It is a figure which shows the structural example of the data which a reputation information storage part stores. It is an example of a screen displayed on a user terminal. It is a functional lineblock diagram of a reputation information extraction device. It is a figure for demonstrating the outline | summary of the process of a reputation information extraction apparatus. It is a flowchart which shows operation | movement of a reputation information extraction apparatus. It is a flowchart which shows operation | movement of a reputation information extraction apparatus. It is a flowchart which shows operation | movement of a reputation information extraction apparatus. It is a figure which shows an example of a text. It is a figure for demonstrating a specific example. It is a figure for demonstrating a specific example.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Review site provision apparatus 2 Reputation information extraction apparatus 3 User terminal 4 Communication network 11 To user terminal interface part 12 Review information storage provision part 13 Reputation information provision part 14 Reputation information extraction apparatus interface part 15 Text storage part 16 Reputation information storage part 21 Interview site providing device interface unit 22 Text acquisition unit 23 Preprocessing unit 24 Dependency analysis unit 25 Phrase pair acquisition unit 26 Phrase pair filtering unit 27 Expression extraction unit 28 Expression filtering unit 29 Data storage unit 30 Dictionary storage unit 31 Reputation information Storage unit 32 Reputation information provider

Claims (8)

An evaluation dictionary storage means for storing an evaluation dictionary and an attribute dictionary storage means for storing an attribute dictionary are provided. By referring to the evaluation dictionary and the attribute dictionary, evaluation expressions and attribute expressions can be obtained from the text stored in the text storage means. A reputation information extraction device having a function of extracting reputation information as a set,
Reading the text stored in the text storage means, performing dependency analysis of sentences included in the text, and extracting a pair of clauses having a dependency relationship from the sentence;
A comparison means for comparing each phrase of the phrase pair with expressions included in the evaluation dictionary and the attribute dictionary;
When it is determined by the comparison means that one of the two clauses constituting the clause pair includes an evaluation expression in the evaluation dictionary and the other clause does not include any expression in any dictionary Determining that the other phrase is an attribute dictionary addition candidate phrase including an attribute expression to be added to the attribute dictionary, and the comparing means includes the one phrase including an attribute expression in the attribute dictionary, and the other phrase If it is determined that the phrase in FIG. 5 does not include any expression in any dictionary, the dictionary additional candidate phrase that determines that the other phrase is an evaluation dictionary addition candidate phrase including an evaluation expression to be added to the evaluation dictionary A determination means;
When a phrase pair including the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause satisfies a predetermined condition based on part of speech, an evaluation expression or attribute expression is obtained from the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause. A reputation information extracting device comprising: a new expression adding means for extracting and adding as a new expression to the evaluation dictionary or the attribute dictionary. The new expression adding means determines whether the part of speech of the last word of the first phrase of two phrases in the phrase pair including the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause satisfies a predetermined condition A phrase pair filtering means for determining
When it is determined by the phrase pair filtering means that at least the part of speech of the last word of the first phrase satisfies a predetermined condition, the new expression adding means includes an evaluation expression included in the evaluation dictionary addition candidate phrase 2. The reputation information extracting apparatus according to claim 1, wherein an attribute expression included in the attribute dictionary addition candidate clause is added to the evaluation dictionary or the attribute dictionary as the new expression. In the case where the first phrase is a phrase including the attribute dictionary addition candidate phrase or the attribute expression in the attribute dictionary, the predetermined condition is that at least the part of speech of the last word is a case particle or a coordinator. That is,
In the case where the first phrase is a phrase including the evaluation dictionary addition candidate phrase or the evaluation expression in the evaluation dictionary, the predetermined condition is a part of speech in which the part of speech of the last word has a utilization form. The reputation information extracting device according to claim 2, wherein the utilization form is a basic form or a body connection.   When the phrase pair satisfies the condition in the phrase pair filtering means, the new expression adding means has a part of speech of the last word of the expression extracted from the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause in advance. 4. The reputation information extracting apparatus according to claim 2, wherein whether or not a predetermined condition is satisfied is determined, and the expression is added to the evaluation dictionary or the attribute dictionary when the condition is satisfied. When the expression is an evaluation expression, the predetermined condition is that the last word of the evaluation expression is a noun that is a general noun, an adjective, or an adjective verb stem,
The reputation information extraction device according to claim 4, wherein, when the expression is an attribute expression, the predetermined condition is that the last word of the attribute expression is a general noun or a proper noun.   In the reputation information extracting device, one of two clauses constituting the clause pair includes an evaluation expression in the evaluation dictionary, and the other clause includes an attribute expression in the attribute dictionary by the comparison unit. When it is determined that the evaluation expression is included, the evaluation expression is extracted from the one phrase, the attribute expression is extracted from the other phrase, and the combination of the evaluation expression and the attribute expression is stored as reputation information in the reputation information storage unit. The reputation information extraction device according to claim 1, further comprising reputation information extraction means. An evaluation dictionary storage means for storing an evaluation dictionary and an attribute dictionary storage means for storing an attribute dictionary are provided. By referring to the evaluation dictionary and the attribute dictionary, evaluation expressions and attribute expressions can be obtained from the text stored in the text storage means. It is a dictionary construction method executed by a reputation information extraction device having a function of extracting reputation information as a set,
A phrase pair extraction step of reading the text stored in the text storage means, performing dependency analysis of the sentence included in the text, and extracting a pair of phrases having a dependency relation from the sentence;
A comparison step of comparing each clause of the clause pair with expressions included in the evaluation dictionary and the attribute dictionary;
When it is determined by the comparison step that one of the two clauses constituting the clause pair includes an evaluation expression in the evaluation dictionary and the other clause does not include an expression in any dictionary Determining that the other phrase is an attribute dictionary addition candidate phrase including an attribute expression to be added to the attribute dictionary, and the one step includes an attribute expression in the attribute dictionary by the comparing step; If it is determined that the phrase in FIG. 5 does not include any expression in any dictionary, the dictionary additional candidate phrase that determines that the other phrase is an evaluation dictionary addition candidate phrase including an evaluation expression to be added to the evaluation dictionary A determination step;
When a phrase pair including the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause satisfies a predetermined condition based on part of speech, an evaluation expression or attribute expression is obtained from the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause. A new expression adding step of extracting and adding as a new expression to the evaluation dictionary or the attribute dictionary. The computer comprises an evaluation dictionary storage means for storing an evaluation dictionary and an attribute dictionary storage means for storing an attribute dictionary, and by referring to the evaluation dictionary and the attribute dictionary, an evaluation expression from the text stored in the text storage means, A program for causing a computer to function as a reputation information extraction device having a function of extracting reputation information that is a set of attribute expressions,
Reading out the text stored in the text storage means, performing dependency analysis of the sentence included in the text, and extracting a pair of clauses having a dependency relation from the sentence;
A comparison means for comparing each clause of the clause pair with expressions included in the evaluation dictionary and the attribute dictionary;
When it is determined by the comparison means that one of the two clauses constituting the clause pair includes an evaluation expression in the evaluation dictionary and the other clause does not include any expression in any dictionary Determining that the other phrase is an attribute dictionary addition candidate phrase including an attribute expression to be added to the attribute dictionary, and the comparing means includes the one phrase including an attribute expression in the attribute dictionary, and the other phrase If it is determined that the phrase in FIG. 5 does not include any expression in any dictionary, the dictionary additional candidate phrase that determines that the other phrase is an evaluation dictionary addition candidate phrase including an evaluation expression to be added to the evaluation dictionary Determination means,
When a phrase pair including the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause satisfies a predetermined condition based on part of speech, an evaluation expression or attribute expression is obtained from the evaluation dictionary addition candidate clause or the attribute dictionary addition candidate clause. New expression adding means for extracting and adding to the evaluation dictionary or the attribute dictionary as a new expression;
Program to function as.

Images