JP2015125570A - Information processing apparatus, control method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information processing apparatus, a control method, and a computer program capable of more precisely analyzing an evaluation object document without troubling a user when analyzing whether the evaluation object document is affirmative, denial, or others in the evaluation.SOLUTION: The information processing apparatus performs morphological analysis and syntactic analysis of an evaluation object document and analyzes the modification of a term which constitutes the document. When the term is included in the evaluation object and when an evaluation word included in the evaluation object and another evaluation word in modification relation with the evaluation word are in cooccurrence relation, the information processing apparatus further obtains change information indicating that the evaluation polarity reserved for the evaluation words should be changed. The information processing apparatus thereafter determines the evaluation polarity of each document on the basis of the change information.

Description

  The present invention relates to an analysis technique for extracting effective information from a document in document analysis processing, and in particular extracts subjective reputation / evaluation, opinion, etc. from digitized document data on the web. The present invention relates to the technology to be presented and its management technology.

  It is an important issue to extract and present necessary information from the vast amount of documents that are created every day, mainly on the web, and being consumed at the same time. In particular, it is useful for business operators to analyze the evaluation status regarding the reputation of the products they handle and the evaluation / opinions on services.

  In order to deal with such problems, various methods for extracting and presenting subjective evaluations such as product reputation, evaluation and opinion from a large amount of digitized document data have been proposed.

  For example, a word having a meaning representing evaluation or reputation (hereinafter referred to as an evaluation word) is registered in a dictionary together with its evaluation value (an attribute value representing a positive or negative meaning / evaluation). There is a method of extracting an evaluation word and calculating an evaluation value quantitatively from its frequency.

  Patent Document 1 discloses a method for registering a predetermined evaluation word in a dictionary, detecting an evaluation word from a document to be analyzed, obtaining an evaluation value, and determining a positive opinion or a negative opinion doing. This method also discloses a method for considering the configuration of a negative sentence and calculating one specific evaluation value from a plurality of words.

  In Patent Literature 2, as in Patent Literature 1, an evaluation value is obtained from a predetermined evaluation word. However, in consideration of the fact that one evaluation word is inverted depending on the evaluation target. As a countermeasure, a dictionary management technique based on user voting is disclosed.

Japanese Patent No. 4796664 Japanese Patent No. 3738011

  However, regarding the reversal of the evaluation value in Patent Document 1, a general negative word (for example, “no”) is considered. For example, when the evaluation value of the evaluation word “disease” is negative, The expression “absent” is affirmative when the evaluation value is inverted, but inversion of the evaluation value for a combination of words other than a negative word is not considered when “the disease is cured”, that is, a word combination.

  In addition, with respect to such an evaluation value inversion problem, Patent Document 1 may solve the problem by using a “binary relationship” that calculates one specific evaluation value from a plurality of words. However, there is no disclosure of countermeasures for cases where a binary relationship exists at the same time as a single evaluation word, such as “disease cures”, and further, such a binary that reverses the evaluation value. There is a new problem of registering and managing huge amounts of relationships.

  Moreover, in the method described in Patent Document 2, the target word corresponding to the evaluation word is used as the basis for the evaluation value reversal, and a positive evaluation is given for “old temple”, while a negative result is given for “old fresh food”. An example of a manual evaluation is disclosed.

  However, the evaluation as in this case is largely due to user voting, and there are cases where the evaluation can be performed without performing user voting in a predetermined combination of words. The technique that should be dealt with in the system without performing user voting is not clearly specified.

  Further, although the same can be said for Patent Document 1, not only the target word but also the word corresponding to the main word can be the basis for the evaluation value to be reversed. For example, an evaluation value of “high price” is a negative evaluation from the viewpoint of the consumer, but a positive evaluation can be considered from the viewpoint of the seller.

  Thus, extracting subjective evaluations such as reputation, evaluation and opinion from document data is an important issue, and it is necessary to calculate evaluation values more efficiently and accurately.

  The present invention has been made in order to solve the above-described problems, and it is possible to analyze whether a document to be evaluated is a positive evaluation or a negative evaluation without taking time and effort to a user. An information processing apparatus, a control method, and a program capable of analyzing an evaluation target document with higher accuracy.

  A first invention for achieving the above object is an information processing apparatus for analyzing affirmative evaluation or negative evaluation in a document, the evaluation word to be evaluated obtained from morphological analysis and syntactic analysis of the document, and the evaluation word Evaluation evaluation for acquiring evaluation word different from the evaluation word which is a dependency relationship between evaluation information acquisition means for acquiring evaluation word information including the evaluation polarity of the evaluation information and evaluation words of the evaluation information acquired by the evaluation information acquisition means A co-occurrence relation determining means for determining whether or not a word acquisition means and an evaluation word that is in a dependency relationship with the evaluation word acquired by the dependency evaluation word acquisition means are in a co-occurrence relationship; And a change information acquisition means for acquiring change information for changing the evaluation polarity of evaluation words in a co-occurrence relationship by a determination means.

  A second invention for achieving the above object is a method of controlling an information processing apparatus that analyzes positive evaluation or negative evaluation in a document, wherein the information processing apparatus is obtained by morphological analysis and syntactic analysis of the document. The evaluation word which is a dependency relationship between the evaluation information acquisition step for acquiring evaluation word information including the evaluation word to be evaluated and the evaluation polarity of the evaluation word, and the evaluation word of the evaluation information acquired by the evaluation information acquisition step; Co-occurrence that determines whether or not a dependency evaluation word acquisition unit that acquires different evaluation words and an evaluation word that is in a dependency relationship with the evaluation word acquired by the dependency evaluation word acquisition unit have a co-occurrence relationship Executing a relationship determination step and a change information acquisition step of acquiring change information for changing the evaluation polarity of the evaluation words in the co-occurrence relationship by the co-occurrence relationship determination step And wherein the door.

  A third invention for achieving the above object is a program readable and executable by an information processing apparatus for analyzing positive evaluation or negative evaluation in a document, wherein the information processing apparatus is used for morphological analysis and syntax analysis of the document. The evaluation information acquisition means for acquiring evaluation word information including the evaluation word to be evaluated and the evaluation polarity of the evaluation word, and the evaluation word of the evaluation information acquired by the evaluation information acquisition means It is determined whether or not the dependency evaluation word acquisition unit that acquires an evaluation word different from the evaluation word and the evaluation word acquired by the dependency evaluation word acquisition unit and the evaluation word that is in a dependency relationship have a co-occurrence relationship. Function as a co-occurrence relation determining means, and a change information acquiring means for acquiring change information for changing the evaluation polarity of evaluation words in a co-occurrence relation by the co-occurrence relation determining means And characterized in that.

  According to the present invention, in analyzing whether a document to be evaluated is a positive evaluation or a negative evaluation, an analysis is performed in consideration of the field attributes of evaluation words constituting the document to be evaluated. Therefore, the evaluation target document can be analyzed with higher accuracy without taking time and effort for the user.

It is a block diagram which shows the structural example of the document analyzer which concerns on embodiment of this invention. It is a block diagram which shows the hardware constitutions of the document analyzer which concerns on embodiment of this invention. It is a flowchart of the document analysis process performed with the document analysis apparatus which concerns on embodiment of this invention. It is explanatory drawing for demonstrating an example of the normalization process in embodiment of this invention. It is a flowchart of the evaluation word extraction process performed with the document analysis apparatus which concerns on embodiment of this invention. It is an example of the duplication candidate deletion process in embodiment of this invention. It is explanatory drawing for demonstrating an example of the co-occurrence process in embodiment of this invention. It is explanatory drawing for demonstrating an example of the dependency relation in embodiment of this invention. It is explanatory drawing for demonstrating an example of the dependency relation in embodiment of this invention. It is a block diagram which shows the structure of the evaluation word candidate list | wrist in embodiment of this invention. It is a flowchart of the co-occurrence process executed by the document analysis apparatus according to the embodiment of the present invention. It is a flowchart of the negative expression process performed with the document analyzer which concerns on embodiment of this invention. It is a flowchart of the speaker information extraction process performed with the document analyzer which concerns on embodiment of this invention. It is a block diagram which shows an example of the structure of the table for memorize | storing the evaluation word extraction result and speaker information extraction result in embodiment of this invention. It is a flowchart of the polarity determination process performed with the document analyzer which concerns on embodiment of this invention. It is an example of the evaluation word dictionary in the embodiment of the present invention. It is an example of the field inversion word in embodiment of this invention. It is an example of the synonym dictionary in the embodiment of the present invention. It is an example of the speaker information dictionary in the embodiment of the present invention.

  Hereinafter, an example of an embodiment of the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram illustrating a configuration of a document analysis apparatus as an information processing apparatus according to an embodiment of the present invention.

  The document analysis apparatus 100 includes an evaluation dictionary unit 101, a synonym dictionary unit 102, a reputation information extraction unit 103, a speaker information extraction unit 104, a polarity determination unit 105, and a speaker information dictionary unit 106. The evaluation dictionary unit 101, the synonym dictionary unit 102, and the speaker information dictionary unit 106 are stored in a storage device such as an external memory 211 described later.

  The document analysis apparatus 100 sends the text document 107 to the reputation information extraction unit 103 and the speaker information extraction unit 104 to extract various information. In each extraction unit, extraction processing is performed with reference to the results of morphological analysis and syntax analysis and information in various dictionaries.

  The extracted results are sent to the polarity determination unit 105, and the evaluation polarity of the text document 107 is calculated. These series of document analysis techniques will be described in detail later.

  Next, the hardware configuration of the document analysis apparatus 100 in FIG. 1 will be described with reference to FIG.

  In the figure, a CPU 201 comprehensively controls each device and controller described later connected to a system bus 204. Further, the ROM 203 or the external memory 211 is used for causing the CPU 201 to execute a basic input / output system (BIOS), an operating system program (hereinafter referred to as an OS), and various processes described later. Various necessary programs and data are stored. The RAM 202 functions as a main memory, work area, and the like for the CPU 201.

  The CPU 201 implements various processes to be described later by loading a program or the like necessary for executing the process into the RAM 202 and executing the program. An input controller (input C) 205 controls input from the input device 209. The input device 209 is configured by, for example, a mechanical keyboard, a software keyboard, a touch panel, or the like. A video controller (VC) 206 controls display on the display device 210. The display device 210 is configured by a liquid crystal display, for example.

  The memory controller (MC) 207 is stored in a hard disk (HD), solid state disk (SSD), or PCMCIA card slot for storing boot programs, browser software, various applications, font data, user files, editing files, various data, and the like. Controls access to an external memory 211 such as a compact flash (registered trademark) memory connected via an adapter.

  A communication I / F controller (communication I / FC) 208 is connected to and communicates with an external device via a network, and executes communication control processing in the network. For example, Internet communication using TCP / IP is possible.

  Note that the CPU 201 enables display on the display device 210 by executing, for example, outline font rasterization processing on a display information area in the RAM 202. The above is a description of the hardware configuration of the document analysis apparatus 100. Needless to say, the hardware configuration illustrated in FIG. 2 is not necessarily required as long as various processes described below can be performed. .

  Next, document analysis processing in the document analysis apparatus 100 will be described in detail with reference to FIGS.

  FIG. 3 is a flowchart showing the overall processing in the document analysis processing. In the document analysis processing, the CPU 201 performs evaluation word extraction processing and speaker information extraction processing after performing morphological analysis and syntax analysis for each sentence described in the document. Thereafter, polarity determination processing is performed based on each extracted data.

  Since the document analysis process is executed in units of sentences, first, in step S301, the document is divided into sentences and the subsequent processes are executed. Sentence splitting is realized by treating punctuation marks, exclamation marks, question marks, continuous newline strings, etc. as delimiters. In subsequent step S302, morphological analysis of the extracted sentence is performed.

  Subsequently, in step S303, normalization processing is performed. The reason for performing the normalization process is to keep the accuracy of the subsequent parsing constant, for example, the process of returning the honorific and polite expression to the standard form, or returning the utilization form to the original form.

  FIG. 4 shows an example of normalization processing for the sentence “I received a delicious fish”. The notation of the sentence 401 is described as a morpheme string obtained from the morpheme analysis result (step S302), and the notation of the sentence 402 is described as a morpheme string obtained from the normalization processing result (step S303). .

  In the sentence 401, since the underlined part “Grate” is a modest word of “eat”, the result of the normalization process in step S303 is corrected as in the sentence 402. In this example, two amendments are made: “Make the modest word a standard form” and “Make the usage form a prototype”. Information for correction is included in the result of the morphological analysis performed in step S302.

  Returning to FIG. 3, syntax analysis is performed in the subsequent step S304. In the syntax analysis, dependency analysis processing for specifying a dependency destination between clauses is performed, and a sentence structure in a phrase unit is acquired.

  Common parsing tools include CaboCha and KNP. In particular, CaboCha is a tool that can perform the morphological analysis of step S302 at a time in the parsing, and therefore the parsing of step S304 instead of the morphological analysis of step S302. After performing the above, it is necessary to perform the normalization process in step S303 based on the information about the morpheme included in the syntax analysis result.

  When step S304 is performed, two analysis results are obtained for the input document: a morphological analysis result and a syntax analysis result. From these analysis results, an evaluation word extraction process in step S305 and a speaker information extraction process in step S306 are performed.

  First, the evaluation word extraction process in step S305 will be described with reference to FIGS.

  The evaluation word extraction process extracts evaluation words registered in advance in the evaluation dictionary unit 101 (the evaluation dictionary shown in FIG. 16) by performing pattern matching on the analysis result obtained from the morphological analysis in step S302. It is processing to do.

  FIG. 16 shows an example of the evaluation dictionary. The evaluation dictionary is a dictionary in which words that can be evaluated expressions are registered. A unique ID value is defined for each word (word ID 1601 in FIG. 16) and is uniquely managed.

  For each word ID 1601, a notation 1602, an initial polarity 1603, a co-occurrence attribute 1604, and a field 1606 are defined. These details are as described above. Note that the word ID 1601 is an ID value managed in the evaluation dictionary by performing an arbitrary calculation on the ID value.

  First, in step S501, the evaluation word candidate list is initialized. The evaluation word candidate list is an area for temporarily storing evaluation words detected in the subsequent extraction process and attached information related to the evaluation words.

  In subsequent steps S502 to S506, the morpheme analysis result (step S302) and the morpheme string obtained by the normalization process (step S303) match the evaluation words registered in the evaluation dictionary unit 101 in advance. It is determined whether or not there is.

  If it is determined that there is a match (if “Yes” in step S504), the process proceeds to step S505, and the word ID, morpheme ID, and phrase ID are added to the evaluation word candidate list in combination and matched. If it is determined that there is no (No in step S504), the process proceeds to step S506 (returns to step S502).

  FIG. 10 shows an example of the evaluation word candidate list. The word ID 1001 stores a word ID for uniquely identifying the extracted word of the evaluation word candidate, and the word ID is managed by the evaluation dictionary unit 101.

  The morpheme ID 1002 is a morpheme identification number in the morpheme analysis result obtained in step S302. Similarly, the phrase ID 1003 is a phrase identification number in the syntax analysis result obtained in step S304.

  Note that the notation 1004 is provided for convenience in order to facilitate understanding of the chart. In this example, the word ID of the evaluation candidate word “price” is 167, which means that the morpheme ID is 4 and the phrase ID is 2 as extracted.

  In the evaluation word search in step S503, it is also possible to search for synonyms and synonyms registered in the synonym dictionary unit 102. The relationship between the evaluation dictionary unit 101 and the synonym dictionary unit 102 will be described later.

  In subsequent step S507, the duplication candidate deletion process is performed. The duplication candidate deletion process will be described with reference to FIG.

  The sentence 601 in FIG. 6 is a result of performing the morphological analysis process in step S302 and the normalization process in step S303 on the sentence “I got mumps”. Here, it is assumed that the evaluation words “muffled cold” and “cold” are registered in the evaluation dictionary unit 101.

  At this time, in the evaluation word search of step S503, both “muffled / cold” consisting of the first morpheme and second morpheme of the sentence 601 and “cold” only of the second morpheme are registered in the evaluation word candidate list.

  As an example of this mode, pattern matching using a plurality of keywords such as the Aho-Corasick method can be cited as an example.

  In the duplication candidate deletion process, when such an arbitrary evaluation word candidate (“cold” in the sentence 601) is included in another evaluation word candidate (“muffled / cold” in the sentence 601), a candidate having a large number of constituent morphemes is selected. Prioritize. In the example of the sentence 601, priority is given to “muffled / cold” composed of two morphemes, and “cold” composed of one morpheme is deleted from the evaluation word candidate list.

  In subsequent step S508, co-occurrence processing is performed. The co-occurrence process will be described with reference to FIGS. 7, 8, 9, 11, and 12.

  The co-occurrence process is a process for detecting an expression representing some evaluation together with other words (hereinafter referred to as an evaluation expression). A sentence 701 in FIG. 7 is a result of performing the morphological analysis process in step S302 for the sentence “This year's Sanma is expensive”. In this example, “price” is not an evaluation expression by itself, but is an example of an evaluation expression when it appears together with “high”.

  FIG. 8 shows a dependency analysis result that is a result of the syntax analysis of the sentence 701. The dependency is a parsing result indicating which clause relates to which clause when the sentence is divided into clauses.

  In FIG. 8, the phrase 801 relates to the phrase 804, the phrase 802 relates to the phrase 803, and the phrase 803 relates to the phrase 804. At this time, focusing on the phrase 803 including “price” and the phrase 804 including “high”, it can be seen that the phrase 803 is related to the phrase 804. Therefore, since it can be determined that “price” and “high” are in a co-occurrence relationship, the evaluation expression (price, high) is effective.

  On the other hand, since “price” and “high” appear in the sentence, it cannot be judged as evaluation expressions. Such word combinations must be in relevant positions in the sentence, as described above.

For example, FIG. 9 shows a syntax analysis result of the sentence “How much is the price of a sama sold by a tall person?”. Since the phrase 906 including “price” relates to the phrase 907 and the phrase 902 including “high” relates to the phrase 903, it can be determined that “price” and “high” do not have a co-occurrence relationship. Thus, in order to determine whether or not two words are related in a sentence, the syntax analysis result obtained in step S304 is used.

  Details of the co-occurrence process will be described with reference to FIG. One evaluation word candidate is extracted from the evaluation word candidate list added in step S505 in FIG. 5, and evaluation word information including information such as the polarity of the word is obtained from the evaluation dictionary unit 101 using the word ID as a key (step S1102). ).

  As will be described later, when the word ID is a synonym, the representative word is retrieved from the synonym dictionary unit 102 (the synonym dictionary shown in FIG. 18), and the evaluation word information of the word is acquired from the evaluation dictionary unit 101. .

  Next, FIG. 18 shows an example of a synonym dictionary. The synonym dictionary manages synonyms and synonyms for various words used in this system, such as evaluation expressions, representative inverted words, or predicates in speaker information.

  A unique ID value is defined for each synonym (synonym ID 1801 in FIG. 18) and is uniquely managed. For each synonym ID 1801, a notation 1802 and a word ID 1803 are set. The word ID 1803 means a link to the unified notation of the synonym. For example, the unified notation of “price”, which is the synonym ID value 10223, has a word ID value of 167, which is understood to be “price” in the evaluation dictionary. Note that the synonym ID 1801 is also an identification number that can be recognized as a synonym by an arbitrary calculation as described above.

  In step S1103, initial polarity is given from the evaluation word information acquired in step S1102. The initial polarity is a polarity set as an initial value in the evaluation word, and an affirmative (positive number), negative (negative number), or neutral (0) polarity is set.

  In the subsequent step S1104, it is determined whether or not a co-occurrence attribute is set in the evaluation word information, and if it is determined that the co-occurrence attribute is not present (a value is not set in the co-occurrence attribute 1604) (in step S1104). If “no”, the process proceeds to step S1111 and if it is determined that the co-occurrence attribute is present (a value is set in the co-occurrence attribute 1604) (“Yes” in step S1104), the process proceeds to step S1105. move on.

  In step S1105, as described above, a process of confirming the related phrase of the evaluation word candidate is performed. It is checked whether another evaluation word candidate having a related phrase ID of the evaluation word candidate is in the evaluation word candidate list.

  Further, the information on the relation clause can be obtained by referring to the syntax analysis result obtained in step S304. Alternatively, the evaluation word candidate list shown in FIG. 10 may be provided with a clause ID indicating a related clause as a syntax analysis result, and information on the related clause may be acquired by referring to this clause ID.

  For example, when the phrase ID of the phrase 803 in FIG. 8 is 2 and the phrase ID of the phrase 804 is 3, the syntax analysis result includes information in which the dependency relationship is established from the phrase ID 2 to the phrase ID 3. 10 together with the example of the evaluation word candidate list shown in FIG. 10, since the phrase ID of the word ID 167 is 2 and the phrase ID of the word ID 938 is 3, the word ID 167 has a dependency relationship with the word ID 938. It can be judged that it is established.

  If there are evaluation word candidates that satisfy such a dependency relationship in the evaluation word candidate list (in the case of “Yes” in step S1106), the process proceeds to step S1107, and the evaluation is such that the dependency relationship is satisfied. If the word candidate does not exist in the evaluation word candidate list (“NO” in step S1106), the process proceeds to step S1111.

In step S1107, it is determined whether the two words having the dependency relationship are co-occurrence relationships. The determination is made by referring to the evaluation word information acquired from the evaluation dictionary unit 101 described above. For example, since “high” is set in the co-occurrence attribute 1604 of the word “price”, it is determined that the co-occurrence relationship is established (in the case of “Yes” in step S1107), and the process proceeds to step S1108. If it is determined that the receiving relationship is established but not the co-occurrence relationship (“No” in step S1107), the process proceeds to step S1111.
In step S1108, an evaluation expression with polarity is detected.

  On the other hand, if “NO” in step S1104, “NO” in step S1106, or “NO” in step S1107, since there is no co-occurrence relationship, etc., is the evaluation word candidate established by the evaluation word candidate alone? Determine if. When it is determined that the initial polarity is neutral (in the case of “No” in step S1111), since there is no polarity, it is not detected as an evaluation expression, so the process proceeds to step S1110, and when it is determined that the initial polarity is not neutral (step In the case of “Yes” in S1111, the process proceeds to step S1108 and is detected as an evaluation expression with polarity.

  For the evaluation expression detected in step S1108, it is determined in subsequent step S1109 whether or not the evaluation polarity is changed by a negative expression. FIG. 12 shows a flowchart of negative expression processing.

  First, in step S1201, field information (field 1606) of the evaluation word is acquired from the evaluation dictionary unit 101. The field information represents the field to which the evaluation word belongs. For example, the words “cancer”, “disease”, and “injury” are defined as belonging to the “medical” field.

  In the field information, a representative inverted word is defined (represented inverted word 1702 among the field inverted words shown in FIG. 17). For example, the representative inverted word of “medical” is defined as “cure”. That is, when an evaluation word related to “medical” is detected, if “cure” appears in the co-occurrence relationship, the object is to reverse the polarity.

  Next, FIG. 17 shows an example of the field inversion word. One representative inversion word 1702 is defined for the field 1701. A word ID 1703 indicates the ID value of the representative inversion word 1702. As described above, the ID value of the representative inversion word can be understood to be a representative inversion word by performing an arbitrary calculation.

  In a succeeding step S1202, it is determined whether or not the representative inverted word described above is included in the relation of the extracted evaluation expression. If it is determined that the evaluation expression does not have a co-occurrence relationship (“No” in step 1104 and “Yes” in step S1111 in FIG. 11), the process advances to step S1204 because there is no dependency destination.

  On the other hand, if it is determined that the co-occurrence relationship is established, the related phrase is confirmed, and if there is the representative inverted word (representative inverted word 1702) (in the case of “Yes” in step S1202), the process proceeds to step S1203. If not (if “NO” in step S1202), the process proceeds to step S1204.

  When a synonym of a representative inverted word is set in the synonym dictionary unit 102, the same processing as that of the representative inverted word is performed. For example, “treat”, “completely cure” (notation 1802) and the like correspond to the synonyms of the representative inverted word “cure”.

  In step S1203, negative information based on representative inverted words is given. Instead of inverting the polarity of the extracted evaluation expression, only the negative information is given, and the final evaluation polarity is determined by the polarity determination unit 105.

  In a succeeding step S1204, it is determined whether or not a negative expression is included in the relation of the extracted evaluation expression. If it is determined that the co-occurrence relationship has been established, check whether the related clause contains, for example, the auxiliary verb “None” or the presence of a negative expression such as “None”. If it is determined that there is no negative expression process, the negative expression process is terminated. If a negative expression is included (in the case of “Yes” in step S1204), the process proceeds to step S1205, and negative information is added in the same manner as in step S1203.

An example of negative information processing is given. In the case of the sentence “I did not get sick”, the result of the parsing is [ Ill / Ne ] → [Become / N / t]
Thus, since the negative auxiliary verb “None” exists at the destination of the evaluation expression “disease”, negative information is given to the initial polarity (−1) of the evaluation expression “disease”. Also, in the case of the sentence “Illness has been cured”, the parsing result is [ Illness / Gas ] → [ Healing / Ta ]
Thus, since the representative inversion word “cure” of the field exists at the destination of the evaluation expression “disease”, negative information is given to the initial polarity (−1) of the evaluation expression “disease”. Also, in the case of the sentence “Disease was not cured”, the result of the parsing is [ Illness / Gas ] → [ Cure / No / Ta ]
Since there is a representative inversion word “cure” in the field of the evaluation expression “disease”, negative information is given to the initial polarity (−1) of the evaluation expression “disease”, and the negative auxiliary verb “no” is further added. Since it exists, further negative information is given. That is, two pieces of negative information are assigned to the initial polarity (−1).

  Returning to FIG. 11, the above process is repeated for all the evaluation word candidates in the evaluation word candidate list, and the co-occurrence process is terminated.

  Returning to FIG. 6, the evaluation word extraction processing is completed by the above processing. At this stage, the evaluation expression, the presence or absence of negative information, the result of parsing, etc. are obtained from the sentence.

  While performing the evaluation word extraction process, the speaker information extraction unit 104 performs speaker information extraction processing. The extraction of the speaker information is to estimate the “position” of the speaker who is the main character of the sentence.

  For example, in the case of the sentence “I did not buy Sanma because the price was high”, the verb information “Buy” is used, so that the speaker information can be estimated as “Consumer”.

  Here, the evaluation expression “price is high” will be described as an example. When the speaker information is “consumer”, it is easily estimated that the evaluation expression “price is high” has a negative polarity. On the other hand, when the speaker information is “supplier”, it is presumed that the polarity is positive.

  In other words, the evaluation expression “price is high” should be judged as negative polarity in the sentence “I didn't buy sanma because the price was high”, and positive in the case of the sentence “yama yesterday was high in price” It is. A method for extracting such speaker information will be described with reference to FIG.

  In step S1301 of FIG. 13, the final phrase is selected from the result of the syntax analysis. In a succeeding step S1302, it is determined whether or not the final clause is a predicate accompanied by a citation clause. A quote clause requires a predicate such as “I think” or “I think”, and the sentence to be evaluated is included in the quote clause.

  If it is determined that a quoted clause is included (“Yes” in step S1302), the process proceeds to step S1303, and a quoted clause is selected instead of the last clause selected in step S1301. Specifically, the citation section can be selected by referring to the dependency analysis result and selecting the clause related to the last section.

  In a succeeding step S1304, it is determined whether or not a speaker precaution is included in the selected phrase. The speaker predicates are, for example, the above-mentioned prescriptions such as “buy” or “sell”, and are registered in the speaker information dictionary unit 106 (speaker information dictionary shown in FIG. 19) as speaker information.

  Next, FIG. 19 shows an example of the speaker information dictionary. Like other dictionaries, the speaker information dictionary also has a speaker information ID 1901 that is a unique ID value, and has a statement 1902, an evaluation expression 1903, a polarity 1904, and a speaker attribute 1905. As described above, the speaker information ID can be determined to be speaker information by an arbitrary calculation.

  If it is determined that the selected phrase contains a speaker precaution (if “Yes” in step S1304), the process proceeds to step S1305, where the speaker information is acquired and it is determined that the speaker prescript is not included (In the case of “No” in step S1304), since there is no speaker information, the speaker information extraction process is terminated.

  Note that the synonym dictionary unit 102 can also be used for searching for the speaker's words. In this case, representative words are acquired from the synonym dictionary and the speaker information is obtained from the speaker information dictionary unit 106. For example, there is a case where “buy” is registered in the synonym dictionary unit 102 when the representative word of the speaker's word is “buy”.

  In a succeeding step S1306, it is determined whether or not there is an inverse connected particle in the phrase including the speaker precaution. Specifically, it is determined whether or not “ga”, “but”, “noni”, and the like are included in the clause, and when it is determined that the connected particle of the reverse connection is included (in the case of “Yes” in step S1306), Proceeding to step S 1307, polarity correction information based on reverse connected particles is added. On the other hand, if it is determined that the connected particle is not included (in the case of “No” in step S1306), the speaker information extraction process is terminated as it is.

  Note that the speaker information is not always specified in the text. Therefore, for example, the result of the previous speaker information extraction process is temporarily stored, and the temporarily stored speaker information is referred only when the speaker information is not described in the subsequent sentence. , And so on.

  If the speaker information is detected at the stage where the speaker information extraction process is completed, the speaker information acquired in step S1305 and the connected particle of the reverse connection are detected, and are added in step S1307. Polarity correction information has been acquired. The polarity determination process in step S307 in FIG. 3 is performed by combining the result of the above speaker information extraction process and the result of the evaluation word extraction process described above.

  FIG. 14 shows an example of the result of the evaluation word extraction process and the result of the speaker information extraction process. When the sentence “Sanma was not bought because the price was high” is processed, the result of the evaluation word extraction process is shown in 1401, and the result of the speaker information extraction process is shown in 1405. These may be configured to include a table in a predetermined storage area and store the result in the table.

  In the evaluation word extraction process, an evaluation expression 1402 and an initial polarity 1404 are extracted. Since the partial character string “because the price was high” including the evaluation expression does not include the representative inversion word or the negative expression, there is no negative information in the result 1401. The word ID 1403 is an ID value of the word “price” registered in the evaluation dictionary unit 101.

  In the speaker information extraction process, the prescription 1407 extracted from the last phrase and the speaker information ID 1406, polarity 1408, and evaluation expression 1409 that are the speaker information are extracted. There is no paradoxical connection particle. The word ID 1410 is an ID value of the word “price” set in the evaluation expression 1409 that is a part of the speaker information.

  Next, the polarity determination process will be described with reference to FIG. The polarity determination process is a process of determining a final evaluation polarity from the evaluation word extraction result and the speaker information extraction result.

  In step S1501 of FIG. 15, it is determined whether or not an extraction result has been obtained by the evaluation word extraction process. If it is determined that there is no extraction result (“No” in step S1501), an evaluation expression for setting the polarity is Since it does not exist, the polarity determination process is terminated, and when it is determined that there is an extraction result (in the case of “Yes” in step S1501), the process proceeds to step S1502.

  In subsequent step S1502, it is determined whether or not an extraction result has been obtained by the speaker extraction process. If it is determined that there is no extraction result (in the case of “No” in step S1502), the process proceeds to step S1505, and there is an extraction result. (Yes in step S1502), the process proceeds to step S1503.

  In step S1503, it is determined whether or not an evaluation expression set for the speaker word obtained as a result of the speaker extraction process exists in the result of the evaluation word extraction process.

  In the example of FIG. 14, since the evaluation expression 1409 “price, high” is obtained by the speaker information extraction process, it is determined whether or not it matches the evaluation expression 1402 in 1401 which is the result of the evaluation word extraction process. .

  If it is determined that the results match (if “Yes” in step S1503), the process proceeds to step S1504. If it is determined that the results do not match (“no” in step S1503), the process proceeds to step 1505.

  In subsequent step S1504, since the two evaluation expressions described above match, the polarity in the speaker information is applied to the evaluation word extraction result. In the example of FIG. 14, the neutrality (0) of the polarity 1404 is replaced with the negation (−1) of the polarity 1408.

  In subsequent steps S1505 to S1507, the negative information detected in the evaluation word extraction process is applied (the processes in steps S1203 and S1205 in FIG. 12).

  In further subsequent step S1508, it is determined whether or not there is polarity information (processing in step S1307 in FIG. 13) due to the reverse connected particle in the speaker information extraction processing. If it is determined that there is polarity information (in the case of “Yes” in step S1508), the process proceeds to step S1509, the polarity is reversed, and if it is determined that there is no polarity information (in the case of “No” in step S1508), polarity correction is performed. The polarity determination process is terminated without performing the process.

  In the example of FIG. 14, as described above, the polarity 1404 is changed from neutral (0) to negative (−1) and no other negative information is given, so the polarity is negative (−1) as a whole. decide. Therefore, the evaluation polarity of the sentence “I didn't buy because of the high price” was negative.

  Next, an example of each dictionary is shown in FIG. 16 to FIG. 19, but all the phrases used in the evaluation dictionary unit 101, the synonym dictionary unit 102, and the speaker information dictionary unit 106 are converted into one pattern matching machine (Tri method). It goes without saying that an efficient extraction process can be carried out in the system or apparatus by registering and processing in (Established etc.).

  As described above, according to the present invention, in analyzing whether a document to be evaluated is a positive evaluation or a negative evaluation, the field attributes of evaluation words constituting the document to be evaluated are considered. Since the analysis can be performed, it is possible to analyze the evaluation target document with higher accuracy without taking time and effort for the user.

  Although the embodiments have been described in detail above, the present invention can take an embodiment as, for example, a system, an apparatus, a method, a program, or a storage medium, and specifically includes a plurality of devices. The present invention may be applied to a system that performs such a process, or may be applied to an apparatus that includes a single device.

  It should be noted that the configuration and contents of the various data described above are not limited to this, and it is needless to say that they are configured with various contents according to applications and purposes.

  Needless to say, the present invention can also be applied to a case where the present invention is achieved by supplying a program to a system or apparatus. In this case, by reading a storage medium storing a program represented by software for achieving the present invention into the system or apparatus, the system or apparatus can enjoy the effects of the present invention.

  Furthermore, by downloading and reading a program represented by software for achieving the present invention from a server, database, etc. on a network using a communication program, the system or apparatus can enjoy the effects of the present invention. It becomes.

  In addition, all the structures which combined each embodiment mentioned above and its modification are also included in this invention.

DESCRIPTION OF SYMBOLS 100 Document analyzer 101 Evaluation dictionary part 102 Synonym dictionary part 103 Reputation information extraction part 104 Speaker information extraction part 105 Polarity determination part 106 Speaker information dictionary part 107 Text document 201 CPU
202 RAM
203 ROM
204 System Bus 205 Input Controller 206 Video Controller 207 Memory Controller 208 Communication I / F (Interface) Controller 209 Input Device 210 Display Device 211 External Memory

Claims (6)

An information processing apparatus that analyzes positive evaluation or negative evaluation in a document,
Evaluation information acquisition means for acquiring evaluation word information including an evaluation word to be evaluated and an evaluation polarity of the evaluation word obtained from morphological analysis and syntax analysis of the document;
Dependency evaluation word acquisition means for acquiring an evaluation word different from the evaluation word that is a dependency relationship with the evaluation word of the evaluation information acquired by the evaluation information acquisition means;
Co-occurrence relation determining means for determining whether or not an evaluation word acquired by the dependency evaluation word acquiring means and an evaluation word having a dependency relation have a co-occurrence relationship;
Change information acquisition means for acquiring change information for changing the evaluation polarity of evaluation words in co-occurrence relation by the co-occurrence relation determination means;
An information processing apparatus comprising: Change determination information storage means for storing change determination information indicating whether or not to change the evaluation polarity for the evaluation words in the co-occurrence relationship;
The information processing apparatus according to claim 1, wherein the change information acquisition unit acquires the change information from the change determination information stored by the change determination information storage unit.   The information processing apparatus according to claim 1, wherein the change information acquisition unit further changes an evaluation polarity of the change information when a dependency destination of the evaluation word is a negative expression. A speaker identification means for identifying a speaker in the document;
The said change information acquisition means changes the said change information of the said evaluation word further by the speaker specified by the said speaker specific means, The change in any one of Claim 1 thru | or 3 characterized by the above-mentioned. Information processing device. A method for controlling an information processing apparatus that analyzes positive evaluation or negative evaluation in a document,
The information processing apparatus includes:
An evaluation information acquisition step for acquiring evaluation word information including an evaluation word to be evaluated and an evaluation polarity of the evaluation word obtained from morphological analysis and syntax analysis of the document;
A dependency evaluation word acquisition step of acquiring an evaluation word different from the evaluation word which is a dependency relationship with the evaluation word of the evaluation information acquired by the evaluation information acquisition step;
A co-occurrence relationship determination step for determining whether or not the evaluation word acquired in the dependency evaluation word acquisition step and the evaluation word in the dependency relationship are in a co-occurrence relationship;
A change information acquisition step for acquiring change information for changing the evaluation polarity of evaluation words in a co-occurrence relationship by the co-occurrence relationship determination step;
A method for controlling an information processing apparatus, characterized by: A program that can be read and executed by an information processing device that analyzes positive evaluation or negative evaluation in a document,
The information processing apparatus;
Evaluation information acquisition means for acquiring evaluation word information including an evaluation word to be evaluated and an evaluation polarity of the evaluation word obtained from morphological analysis and syntax analysis of the document;
Dependency evaluation word acquisition means for acquiring an evaluation word different from the evaluation word that is a dependency relationship with the evaluation word of the evaluation information acquired by the evaluation information acquisition means;
Co-occurrence relation determining means for determining whether or not an evaluation word acquired by the dependency evaluation word acquiring means and an evaluation word having a dependency relation have a co-occurrence relationship;
Change information acquisition means for acquiring change information for changing the evaluation polarity of evaluation words in co-occurrence relation by the co-occurrence relation determination means;
A program characterized by making it function.

Images