JP2014235511A - Information processing device, control method, and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information processing device, a control method, and a computer program, which allow a user to accurately create a summary text according to a field of a text.SOLUTION: By extracting a feature of each of sentences composing an e-mail from a syntactic structure, with respect to each of the sentences, determining an attribute of each of the sentences, extracting a phrase which may be dependent on a field of a text, and creating a summary text, and then associating its feature with the summary text, it becomes possible to create the summary text corresponding to the feature (field).

Description

  The present invention relates to a technique for extracting effective information from a text such as a text of an e-mail so that the text such as an e-mail can be easily grasped.

  In the e-mail system, in addition to browsing as general text, it is possible to manage and browse e-mail exchanged between senders and receivers in units of threads, even if there are only a few e-mails that make up the thread. For example, it is easy for a user to understand whether a topic in a thread is ongoing or completed, but if there are dozens of emails, individual emails can be sent to understand the topic flow. Therefore, posting a summary sentence or important sentence showing the topic status can be a useful solution.

  Various techniques have been proposed for such problems in order to realize efficient browsing of electronic mail by users or from the viewpoint of document management.

  For example, priority is given to a technique that extracts important words based on statistical information and presents peripheral sentences including the important words, a technique that simply displays a few lines at the top without analysis, or a sentence that includes a specific keyword There is a method to present it automatically.

  However, important words based on statistical information do not necessarily include a sentence indicating the requirement, and the requirement that the sender wants to convey is not necessarily described at the beginning of the email body. There are problems such as that may not function properly depending on the field.

  Therefore, a technique is disclosed in which an input sentence is divided according to a predetermined division rule, and a summary sentence is created based on the presence or absence of an important word included in the sentence after determining the type of the sentence (for example, Patent Document 1). reference).

  In this method, the input sentence is limited to text that includes matters related to the use of the device, and the division rules and the important words associated therewith are also determined in advance, so it is easy to effectively generate summary sentences centered on sentences that can be considered important. .

JP 2012-203460 A

  However, if an attempt is made to apply the technique disclosed in Patent Document 1 to an e-mail whose field cannot be specified, a key word corresponding to the field cannot be defined in advance. It is a general solution to determine important words based on target information.

  However, even in this case, if a sentence of a certain scale is not accumulated, the credibility as an important word based on statistical information becomes poor.

  In this way, it is possible to generate an effective summary sentence and present it to the user even in an environment where the field cannot be identified in advance or the accumulated important sentences cannot be defined in advance, so that it is not possible to define an accurate important word in advance. And efficient document management is necessary.

  The present invention has been made to solve the above problems, and provides an information processing apparatus, a control method, and a computer program that enable a user to accurately create a summary sentence according to the field of text. With the goal.

  A first invention for achieving the above object is an information processing apparatus for creating a summary sentence of a sentence according to a field, and after performing a morphological analysis of a sentence, for a phrase obtained by the morphological analysis Analysis means for analyzing the structure of the dependency, feature identification means for specifying the feature of the sentence using the analysis result by the analysis means, and specifying the content of the sentence using the analysis result by the analysis means A summary sentence of a sentence corresponding to a field is created by using a content extracting means for extracting consecutive phrases and a feature of the sentence specified by the feature specifying means and a continuous phrase extracted by the content extracting means. And a summary sentence creating means.

  A second invention for achieving the above object is a method of controlling an information processing apparatus for creating a summary sentence of a sentence according to a field, obtained after performing a morphological analysis of a sentence, and obtained by the morpheme analysis An analysis step for analyzing the structure of a dependency on a phrase, a feature specifying step for specifying a feature of the sentence using an analysis result of the analysis step, and an analysis result of the analysis step, By using a content extraction step for extracting continuous phrases specifying content, a feature of the sentence specified by the feature specifying step, and a continuous phrase extracted by the content extraction step, a sentence summary sentence according to the field And a summary sentence creating step for creating.

  A third invention for achieving the above object is a program that can be read and executed in an information processing apparatus that creates a summary sentence of a sentence according to a field, and the information processing apparatus performs a morphological analysis of the sentence Thereafter, an analysis means for analyzing a structure of a dependency on a clause obtained by the morphological analysis, a feature specifying means for specifying the feature of the sentence from an analysis result by the analysis means, and an analysis result by the analysis means By using the content extracting means for extracting continuous phrases specifying the content of the sentence, the feature of the sentences specified by the feature specifying means and the continuous phrases extracted by the content extracting means, It is characterized by functioning as a summary sentence creating means for creating a summary sentence of the corresponding sentence.

  According to the present invention, by associating the feature amount obtained by analyzing a sentence with a summary sentence created by extracting a phrase that can depend on the field of the sentence, the user can respond to the field of the sentence. It is possible to create a summary sentence with high accuracy.

It is a figure which shows the structural example of the electronic mail analyzer in embodiment of this invention. It is a figure which shows the hardware constitutions of the electronic mail analyzer in embodiment of this invention. It is a flowchart of the email analysis process in the embodiment of the present invention. It is an example of the electronic mail in embodiment of this invention. It is an example of the original text correction process in embodiment of this invention. It is an example of the dependency analysis process in the embodiment of the present invention. It is an example of the summary sentence production | generation in embodiment of this invention. It is a flowchart of the triple extraction process in the embodiment of the present invention. It is a flowchart of the predicate analysis process in the embodiment of the present invention. It is a flowchart of the principal part analysis process in embodiment of this invention. It is an example of the auxiliary | assistant attribute in embodiment of this invention. It is a flowchart of the content word extraction process in embodiment of this invention. It is a flowchart of the summary sentence production | generation process in embodiment of this invention. It is a block diagram which shows the structure of the mail summary list screen in embodiment of this invention. It is a block diagram which shows the structure of the table used by each process in embodiment of this invention.

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

  FIG. 1 is a diagram showing a configuration of an electronic mail analysis apparatus according to an embodiment of the present invention.

  The electronic mail analysis apparatus 100 includes a mail reception unit 101, a dictionary unit 102, an analysis unit 103, a mail storage unit 104, and a display unit 105. The dictionary unit 102 and the mail storage unit 104 are external memory 211 described later. Is stored in a storage device.

  The mail receiving unit 101 receives an e-mail from the wide area network 110 and stores the received e-mail in the mail storage unit 104. Note that the mail storage unit 104 includes the basic mail information table shown in FIG. 15, and when storing, the ID is used to uniquely identify an email.

  The analysis unit 103 retrieves the email stored in the email storage unit 104, analyzes the email while retrieving data necessary for morphological analysis and syntax analysis from the dictionary unit 102, and displays the analysis result on the display unit 105. To display. The series of email analysis methods will be described later in detail.

  Next, the hardware configuration of the electronic mail analysis apparatus 100 of 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 to execute a BIOS (Basic Input / Output System) or an operating system program (hereinafter referred to as an OS), which is a control program of the CPU 201, or various types of processing described later. Various programs and data necessary for the storage 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 the description of the hardware configuration of the electronic mail analyzing apparatus 100. Needless to say, the hardware configuration shown in FIG. 2 is not necessarily required as long as various processes described below can be performed. Yes.

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

  FIG. 3 is a diagram showing an overall flowchart of the e-mail analysis process. In the e-mail analysis process, the CPU 201 reads the e-mail from the e-mail storage unit 104, performs morphological analysis and syntactic analysis on each sentence described in the e-mail, and then performs a triple extraction process and a content word extraction process. The summary sentence generation process is performed. Thereafter, a labeling process is performed based on each extracted data, and a process for determining a representative summary sentence is performed.

  Since the e-mail analysis process is executed in units of sentences, first, in step S301, the e-mail 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 step S302, morphological analysis is performed on each sentence divided in step S301.

  In step S303, original text correction processing is performed. The original text correction process is a normalization process for keeping the accuracy of the subsequent parsing constant, and includes, for example, a process of returning honorific and polite expressions to a standard form. FIG. 5 shows an example of the original text correction process.

  In the sentence 501 of FIG. 5, the underlined part “just” is a modest word of “get”, so that the sentence is corrected as the sentence 502 as a result of the original sentence correction process. Note that the information for correction is included in the result of the morphological analysis performed in step S302.

  The information for correction included in the result of morphological analysis corresponds to the language that is the target of the text correction process using a table for conversion to a standard language that excludes respect components from honorific expressions and polite expressions Corresponds to the standard language. This point can be realized by existing morphological analysis.

  Returning to FIG. 3, syntax analysis is performed in the subsequent step S304. Here, a process called dependency analysis is performed to specify a dependency destination between clauses. An example of dependency analysis is shown in FIG.

  FIG. 6 shows the dependency analysis result of the sentence 502 in FIG. The phrase 601 relates to the phrase 602, and the phrase 602 indicates that it relates to the phrase 606. Further, the phrase 603 relates to the phrase 604, and the phrase 604 and the phrase 605 indicate that it relates to the phrase 606.

Non-Patent Document 1 shown below defines that dependency analysis has the following four characteristics.
“Non-Patent Document 1” Kiyotaka Uchimoto et al., “Japanese dependency analysis using a model based on the maximum entropy method”, Transactions of the Information Processing Society of Japan, Vol.40, No.9, pp.3397-3407, 1999
(1) Backward modification (2) Non-intersection condition (3) Single receiving element (4) In many cases, a forward context is not required for determining the destination

  The dependency analysis having such characteristics can be output by a general syntax analysis tool such as CaboCha or KNP. In particular, CaboCha is a tool that can perform dependency analysis at once, including morphological analysis in step S302, in syntax analysis. Therefore, after performing syntax analysis in step S304 instead of morphological analysis in step S302, syntax analysis is performed. Based on the morpheme information included in the result, it is necessary to perform the original text correction process in step S303.

  Returning to FIG. 3, at the time when the syntax analysis is performed in step S304, two analysis results of the morphological analysis result and the syntax analysis result (dependency) are obtained for the input sentence. Here, from these analysis results, a triple extraction process in step S305, a content word extraction process in step S306, and a summary sentence generation process in step S307 are performed.

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

  The triplet is defined as a combination of the three features of the “main word”, “word form”, and “auxiliary attribute” of the sentence. Note that the definitions of the features “main words” and “word forms” are the same as those defined in Non-Patent Document 1 described above, and the “auxiliary attribute” is an expression that is supplementarily given to the predicate of the sentence, for example, It is defined as an attribute of expression such as negative expression (not) or guess expression (~).

  As an example, the triplet extracted from the sentence “Train is delayed” is (Train, Delayed, φ), and if there is no corresponding “auxiliary attribute”, the triple is Specify φ indicating the empty attribute.

  In addition, the triple extracted from the sentence “Train was not delayed” includes negative expressions, so (Train, Delayed, Denied). Similarly, the sentence “Trains will be delayed” is a guess expression. Because it includes (train, delay, guess).

  That is, the first term and the second term of the triple are extracted from the morpheme of the sentence and the standard form is designated, and the auxiliary attribute class to which the morpheme string of the sentence belongs is designated as the third term. For the third term, multiple classes may be specified (eg, there will be no → denial / inference).

  FIG. 8 shows a flowchart of the triple extraction process. In step S801, a negative variable is initialized. A negative variable is a variable that counts when a negative expression is detected.

  Here, the reason for counting negative expressions is to appropriately handle prefixes that mean double negation or negation. As will be described later, in the triple extraction process, since the analysis is divided into the main part and the predicate, a common variable is required.

  As an example, the sentence “This story is not unrealistic” is equivalent to the sentence “This story is realistic” because the prefix “non” and the auxiliary verb “none” both mean negation. Need to be considered. In this case, it is better to express the labeling correctly from the viewpoint of data normalization in the subsequent labeling process by expressing the triplet as (spoken, unrealistic, not) (speaking, realistic, φ). It is expected to be possible.

  In step S802, predicate analysis processing is performed. FIG. 9 shows a flowchart of predicate analysis processing.

  In the predicate analysis process, in step S901, the clause determined as the last clause from the syntax analysis result obtained in step S304 is selected.

  In step S902, it is checked whether a negative expression is included in the selected final clause. The presence or absence of negative expression is determined using a grammatical rule consisting of a specific part of speech and notation. For example, it is defined that a morpheme having a grammatical rule “part of speech (auxiliary verb) and notation (not)” is included in the clause, and determination is performed using information stored in the dictionary unit 102 of FIG. Hereinafter, it is assumed that similar rules are stored in the dictionary unit 102.

  If a negative expression is included (in the case of “Yes” in step S902), the process proceeds to step S903. In step S903, the negative variable is incremented by one, and a negative expression is not included (in the case of “No” in step S902). Then, the process proceeds to step S904.

  In step S904, it is determined whether the last clause described above is accompanied by a quote clause. For example, when the last section is “I think” or “I think”, it is determined that the quotation section is accompanied.

  If there is no quotation section (“No” in step S904), the process proceeds to step S908. If a quotation section is included (“Yes” in step S904), the process proceeds to step S905.

  In step S905, a citation section is selected instead of the last section, and in steps S906 and S907, processing equivalent to the processing performed in steps S902 and S903 is performed on the citation section.

  In step S908, the finally selected clause (final clause) is extracted as a predicate clause.

  In step S909, the auxiliary attribute of the predicate clause is determined. FIG. 11 shows an example of auxiliary attributes. As described above, if the predicate clause includes an expression as shown in FIG. 11, the attribute is selected as an auxiliary attribute. Since the list of auxiliary attributes does not depend on the text field in particular (because it is an invariable part as Japanese), a pattern matching dictionary defined in advance by hand can be prepared.

  This completes the predicate analysis processing, and returns to FIG. 8 and proceeds to step S803. In step S803, the standard form of the clause selected in step S908 is set in the triple predicate, that is, the second term.

  Subsequently, in step S804, main part analysis processing is performed. In the main part analysis process, the part that becomes the subject of the predicate described above is specified. The main part analysis process will be described with reference to FIG.

  FIG. 10 shows a flowchart of main part analysis processing. The main part analysis processing is intended to identify in which clause the main part is included from the syntax analysis result.

  First, in step S1001, the predicate clause determined in step S908 described above is selected. The clause preceding the predicate clause selected in the subsequent step S1002 is selected regardless of the dependency structure.

  In step S1003, if the previous section does not exist (in the case of “No” in step S1003), the main part analysis is terminated. In this case, the sentence does not have a main part corresponding to the predicate (for example, an instruction form). If the previous section exists (if “Yes” in step S1003), the process proceeds to step S1004.

  In step S1004, when the clause is a clause related to the clause of the predicate selected in step S1001 (when determined as a dependency destination in the dependency structure), the process proceeds to step S1005, and the clause of the predicate clause is determined. If there is no clause (when the dependency structure is not a dependency destination), the process returns to step S1002, and the previous paragraph is selected regardless of the dependency structure, and steps S1003 and S1004 are repeated. In this way, the search is performed from the back to the front of the sentence in units of clauses, and the main part corresponding to the predicate is searched by determining only whether the relevant phrase relates to the predicate.

  In step S1005 (when it is determined in step S1004 that the clause is related to the predicate), it is determined whether the clause ends with a case particle. If it ends with a case particle (in the case of “Yes” in step S1005), the process proceeds to step S1007. If it does not end with a case particle (in the case of “No” in step S1005), the process proceeds to step S1006. Proceed.

  In step S1006, it is determined whether or not the phrase ends with the particle “ha”. If it is not ended with the particle “ha”, the process returns to step S1002 to further search forward for the phrase as described above. When it ends with the particle “ha” (in the case of “yes” at step S1006), it is regarded as equivalent to the case with the case particle in step S1005, and the process proceeds to step S1007.

  In step S1007 (when it is determined in step S1005 that the clause ends with a case particle, or in step S1006 it is determined that the clause ends with a particle “ha”), the main part of the clause is present. The main part is extracted. The main part is defined by a grammatical rule of a continuous morpheme sequence consisting of the immediately preceding noun, proper noun, compound word, prefix, etc. of the case particle or the particle “ha” detected in the clause.

  In step S1008, it is determined whether a negative prefix is included in the extracted main part. The negative prefix corresponds to “non” or “not yet”. If the prefix is included (in the case of “Yes” in step S1008), the process proceeds to step S1009, the negative variable is increased, the main part extraction process is terminated, and the prefix is not included ( In the case of “No” in step S1008), the main part analysis processing is ended as it is.

  Returning to FIG. 8, the process proceeds to step S <b> 805, and the main part extracted in the main part analysis process is set as a first term of a triple.

  In step S806, it is determined whether a negative expression is established. If the negative expression is an odd number of times through the predicate analysis process and the main part analysis process described above, it is considered that the negative expression is satisfied (in the case of “Yes” in step S806), and the process proceeds to step S807. In step S807, The auxiliary attribute “deny” is set in the third term of the triple, and the process proceeds to step S808. If the negative expression is not satisfied (in the case of “no” in step S806), the process proceeds to step S808 as it is.

  In step S808, it is confirmed whether another auxiliary attribute is extracted in step S909 of FIG. When the auxiliary attribute has been extracted (in the case of “Yes” in step S909), the process proceeds to step S809, the auxiliary attribute of the third term of the triple is updated, and the triple extraction process is terminated. If the auxiliary attribute has not been extracted (“No” in step S909), the triple extraction process is terminated.

  Returning to FIG. 3, the content word extraction processing in step S306 will be described with reference to FIGS.

  FIG. 12 is a flowchart showing content word extraction processing.

  The content word indicates a continuous morpheme string centering on a noun phrase described in the sentence, and means a phrase indicating the contents of the sentence.

  First, in step S1201 to step S1203, content words are extracted from all phrases. At this time, only the content words included in which clauses are related without considering the dependency structure.

  The position of the phrase for association is managed by a phrase number, and the set of the phrase number and the content word is defined as a first content word set. The first term of the first content word set indicates the phrase number, and the second term indicates the extracted content word.

  In the process of extracting content words in step S1202, a prohibited word dictionary is prepared in advance, and words included in the dictionary are not extracted as content words. For example, specific words that do not depend on the field of text, such as directive words such as “this”, “that”, “it”, and words with abstract meanings such as “where” “thing” “thing”. It is registered centering on words and phrases that do not indicate an unusual event.

  An example of the first content word set is given using FIG. If the content word is defined as a noun, compound word, proper noun, adjective, or an adjective verb and the particle "no" in a continuous morpheme sequence, the sentence "next / meeting, but / the following / schedule / time The first content word group of (1, the next meeting) and (3, the following schedule) and (5, time) are extracted as the first content word group.

  Returning to FIG. 12, the process proceeds to step S1204, and the process of outputting the first content word set according to the dependency structure is performed from step S1204 to step S1210.

  Output according to the dependency structure means that when the dependency structure is regarded as a tree structure, a search is performed from the leaf toward the root. In the dependency structure in FIG. 6 described above, the clauses indicating leaves are the clause 601, the clause 603 and the clause 605, and the clause indicating the root is the clause 606 which is the final clause.

  In step S1204, the process starts from a leaf phrase. In step S1205, the second content word set is initialized. The second content word set is a set of content words obtained when a phrase is transitioned according to the dependency structure, and is obtained from the first content word. In step S1205, the second content word set is emptied.

  In step S1206, it is confirmed whether the clause is the final clause. If it is the last phrase (“Yes” in step S1206), the process proceeds to step S1209. If it is not the last phrase (“No” in step S1206), the process proceeds to step S1207. For one content word set having the clause number in the first term, the content word described in the second term of the first content word set is added to the second content word set.

  In step S1208, the transition from the clause to the dependency clause is made, and the processing returns to step S1206, and the content words included in the clause are similarly added to the second content word set.

  In the case of “Yes” in step S1206, that is, when the transition is made to the last phrase, the process proceeds to step S1209, the second content word set is output, and in step S1210, the process is repeated for all remaining leaf phrases. Then, the process returns to step S1204. When the analysis is completed for all leaf clauses, the content word extraction process is terminated.

  An example of the second content word set will be given with reference to FIG. When the second content word group is extracted from the first content word group according to the dependency structure, three second content word groups (next meeting), (the following schedule), and (time) are obtained.

  By outputting the content words as a set according to the dependency structure in this way, it is possible to suppress the extraction of a plurality of content words as a co-occurrence relationship between clauses that are not actually related, such as parallel descriptions. And an analysis result with higher accuracy can be obtained.

  Returning to FIG. 3, the summary sentence generation processing in step S307 will be described with reference to FIGS.

  FIG. 13 is a flowchart showing summary sentence generation processing.

  In step S1301, the area for storing the output is initialized, and in step S1302, the phrase number of the last phrase is added to the storage area. Of course, the phrase character string of the phrase may be added, but it is sufficient that the order (or position information) with respect to the original sentence is clear.

  In step S1303, it is confirmed whether or not the last phrase is accompanied by a quotation. The criterion for determining the quotation is described in the triple extraction process (step S904).

  If there is no quote section (“No” in step S1303), the process proceeds to step S1305. If a quote section is included (“Yes” in step S1303), the process proceeds to step S1304. Then, the clause number of the previous clause of the last clause which is a quote clause is added to the storage area.

  In the processing from step S1305 to step S1308, it is checked whether or not all the clauses before the last added clause in the storage area are clauses related to the last added clause.

  That is, when the final clause is accompanied by a quoted clause, the clause related to the clause immediately before the final clause is targeted, and when the final clause is not accompanied by a quoted clause, the clause related to the final clause is targeted.

  In step S1306, if the dependency destination of the clause is not recorded in the storage area (if “NO” in step S1306), the process proceeds to step S1308, and the dependency destination of the clause is recorded in the storage area. If so (if “Yes” in step S1306), the process proceeds to step S1307, and in step S1307, the clause is added to the storage area.

  In step S1309, the phrase added to the storage area is detected including a part of the content word obtained by the content word extraction process described above. If the phrase has a part of the content word (“Yes” in step S1310), the process proceeds to step S1311. In step S1311, the previous phrase related to the phrase is stored so as to completely include the content word. If it is added to the area and otherwise (“No” in step S1310), the process proceeds to step S1312.

  In step S1313, the phrases added to the storage area are arranged in the order of appearance positions to generate a summary sentence. If only the clause numbers are stored, the summary numbers may be generated by arranging the clause numbers and using the clause character strings corresponding to the clause numbers.

  An example of summary sentence generation will be described with reference to FIG.

  A sentence having a structure as shown in FIG. 7 as a dependency structure will be described as an example, “I'm sorry for the convenience of this, but I would like to make it the end of the year by a meeting in November”.

  First, the final phrase 709 “I think” is stored in the storage area. Since “I think” is accompanied by a quotation, the immediately preceding phrase 708 “I want to” is also saved in the storage area.

  Next, for all the clauses before the clause 708, those related to the clause 708 are added. The phrase 703 “I am sorry,” the phrase 705 “In a meeting,” and the phrase 707 “Finally” correspond.

  Further, the remaining clauses are divided into content words, and some of them are stored in the storage area because the clause 704 “November” and the clause 705 “Meeting” , Add a clause 704. The reason is that the phrase 704 “November” and the phrase 705 “In a meeting” are nouns, compound words, proper nouns, adjectives, or a morpheme string in which the adjective verb and the noun “no” are continuous This is because it is an example of definition.

  As a result of the above processing, a summary sentence “Sorry, I would like to make it last in the meeting in November” is generated.

  Returning to FIG. 3, the processes from the morphological analysis in step S302 to the summary sentence generation in step S307 are performed for all sentences.

  In step S309, it is labeled which configuration each sentence of the e-mail bears.

  This is because e-mail cannot be identified in advance, but the structure of the text is often similar. Especially in e-mail used in business, the description of the recipient, the description of the sender, the greeting, the main subject, the conclusion It is often seen that it has a structure such as a sentence and a signature.

  Therefore, the triplet extracted from each sentence, and the second content, in order to determine “where the composition is changed (separated)” and “which part is the composition” By grasping the word set, the morpheme information, and the like as the feature amount of each sentence, it can be determined using an existing conditional random field model.

  In this step, information on the summary created in each step is stored in the mail summary information table (FIG. 15) provided in the mail storage unit 104.

  The mail summary information table includes a mail ID corresponding to the mail ID in the mail basic information table, a branch number for the mail ID, a triple item for storing each item extracted by the triple extraction process, a labeled label, a summary It includes a mail text (summary original text) that represents the text of the mail before summarization.

  In the mail summary information table, when there are a plurality of sentences for one electronic mail, information regarding the summary is generated for each sentence, so that each record can be stored. Branch numbers are used to uniquely identify each sentence.

  In step S310, a representative summary sentence is selected. A sentence that is determined to be a “main subject” part in the labeling process is selected as a target. At this time, the selection priority of the summary sentence is manipulated by referring to the second and third terms of the triple of the target sentence.

  The selection priority will be described with reference to FIG.

  FIG. 4 shows an example of electronic mail. When the main subject 402 is determined in the electronic mail 401, a sentence 403, a sentence 404, and a sentence 405 constituting the content are considered. A triple obtained from each sentence is a triple 406, a triple 407, and a triple 408.

  At this time, the selection priority can be determined from features such as a triplet or a second content word set. For example, it is possible to determine which sentence is selected by determining the selection priority in advance according to the third term of the triple. Such priorities can be determined manually in advance, or can be prepared as learning data using machine learning or statistical techniques.

  In the example shown in FIG. 4, when a high priority is set for a “question” that requires an answer, the sentence 403 is selected and the representative summary sentence “This is the next meeting. Can you give me time? " By using such a selection priority, there is an effect that the representative summary sentence can be correctly selected even if the order of the sentences is changed or the text of the electronic mail is not neatly shaped.

  In addition, there is no need to select only one representative summary sentence. The method of selecting with reference to the order of priority of selection and the calculated score, etc. A method of selecting and displaying the specified date and time by determining the triplet of the specified date and time lines as “date and time” may be applied.

  Furthermore, the information regarding the summary of the electronic mail obtained in the present invention can be displayed on the mail summary list screen 1400 shown in FIG.

  The mail summary list screen 1400 includes a summary list display area 1401 for displaying information related to mail summaries, a label list 1402 for narrowing down and displaying sentences summarized using labels in the summary list display area 1401, The summary list display area 1401 is provided with a selection priority list 1403 for narrowing and displaying sentences summarized using the selection priority.

  When the user wants to narrow down and refer to information related to the summary displayed in the summary list display area 1401, for example, when narrowing down using the selection priority, the selection priority list using the input device 209 is used. When the user presses 1403, the set selection priority is displayed in a list, and by pressing the desired selection priority using the input device 209, it is possible to narrow down and display information about the summary.

  The mail summary list screen 1400 is displayed using information related to the summary stored in the mail summary information table. Each item name of the mail summary information table and the item name of the mail summary list screen 1400 basically correspond to each other. However, the triple item of the mail summary information table corresponds to the selection priority of the mail summary list screen 1400, and even if all items of the triple item are displayed as the selection priority, respectively. It is also possible to display an arbitrary number of items from the triple item. In the present embodiment, the auxiliary attribute of the third item of the triplet item is displayed as the selection priority.

  In addition, when the user wants to refer to a summary sentence on a screen (not shown) in which basic information of the e-mail stored in the e-mail basic information table is displayed, the e-mail is selected using the input device 209. Or, by pressing a button (not shown) for inputting an instruction to refer to the summary text in the selected state, the mail summary list screen 1400 is displayed, and the summary text regarding the selected electronic mail is displayed in the summary list display area. 1401 can be displayed. In this case, information related to the summary stored in the email summary information table is acquired using the email ID of the selected electronic mail and displayed on the email summary list screen 1400.

  As described above, according to the present invention, by associating the feature amount obtained by analyzing the sentence with the summary sentence created by extracting the words that can depend on the field of the sentence, the user can A summary sentence according to the field can be created with high accuracy.

  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 E-mail analyzer 101 Mail receiving part 102 Dictionary part 103 Analysis part 104 Mail storage part 105 Display part 110 Wide area network 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 (3)

An information processing apparatus for creating a summary sentence of a sentence according to a field,
After performing morphological analysis of the sentence, analysis means for analyzing the structure of the dependency on the clause obtained by the morphological analysis;
Using the analysis result by the analysis means, the feature specifying means for specifying the feature of the sentence;
Content extraction means for extracting continuous phrases that specify the content of the sentence using the analysis result by the analysis means;
A summary sentence creating means for creating a summary sentence of a sentence according to a field by using the feature of the sentence specified by the feature specifying means and a continuous phrase extracted by the content extracting means;
An information processing apparatus comprising: A method of controlling an information processing apparatus that creates a summary sentence of a sentence according to a field,
After performing a morphological analysis of the sentence, an analysis step for analyzing the structure of the dependency on the clause obtained by the morphological analysis;
Using the analysis result of the analysis step, a feature specifying step for specifying the feature of the sentence;
Using the analysis result of the analysis step, a content extraction step for extracting continuous phrases that specify the content of the sentence;
A summary sentence creating step for creating a summary sentence of a sentence according to a field by using a feature of the sentence identified by the feature identifying step and a continuous phrase extracted by the content extracting 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 creates a summary sentence of a sentence according to a field,
The information processing apparatus;
After performing morphological analysis of the sentence, analysis means for analyzing the structure of the dependency on the clause obtained by the morphological analysis;
A feature specifying means for specifying the feature of the sentence from the analysis result by the analysis means;
Content extraction means for extracting continuous phrases that specify the content of the sentence using the analysis result by the analysis means;
A summary sentence creating means for creating a summary sentence of a sentence according to a field by using the feature of the sentence specified by the feature specifying means and a continuous phrase extracted by the content extracting means;
A program characterized by making it function.

Images