JP2012230614A - Information processor, information processing method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To flexibly convert an ambiguous word.SOLUTION: An ambiguous word detection part is configured to detect an ambiguous word whose level is ambiguous from input information to be input from a user, and a task determination part is configured to determine a task relating to a modifiee to which the ambiguous word is related from the input information, and a significance determination part is configured to determine the significance of the task, and a conversion part is configured to apply the ambiguous word as well as the significance to a discriminator as an input to convert the ambiguous word into a concrete numerical value. The discriminator is configured to input the significance of the task and the ambiguous word, and to output the concrete numerical value corresponding to the level expressed by the ambiguous word. This invention is applicable to such a case that the ambiguous word is converted into the concrete numerical value.

Description

  The present technology relates to an information processing device, an information processing method, and a program. An information processing device, an information processing method, and an information processing device that can easily prevent miscommunication by flexibly converting ambiguous words. , Related to the program.

  There has been proposed a dialogue support apparatus that prevents miscommunication caused by the inclusion of ambiguous words with vague degrees in dialogue between users and supports smooth communication (see, for example, Patent Document 1). .

  In the conventional dialogue support apparatus, a conversion table is prepared in which for each of the interlocutors A and B who perform dialogue, an ambiguous word is associated with a range of specific numerical values represented by the ambiguous word.

  Further, in the conventional dialogue support apparatus, an ambiguous word is detected from the utterance of the dialogue person A or B, and referring to the conversion table, a range of specific numerical values assumed by the dialogue person A for the ambiguous word, and the dialogue The range of specific numerical values assumed by the person B for the ambiguous word is detected.

  And in the conventional dialogue support device, there is a difference between the range of specific numerical values assumed by the conversation person A for ambiguous words and the range of specific numerical values assumed by the conversation person B for ambiguous words. A warning message is issued because there is a risk of misunderstanding (miscommunication) between the interlocutors A and B.

JP 2008-083753 A

  In a conventional dialogue support apparatus, when attention is paid to a certain conversation person, the phrase including the ambiguous word is referred to the conversion table of the noticed conversation person, and the specific degree to which the ambiguous word included in the phrase assumed by the attention conversation person is represented. A numerical value (range) is obtained.

  Therefore, in the conventional dialogue support apparatus, the same specific numerical value is calculated | required with respect to the same phrase about an attention dialogue person.

  However, even if it is an ambiguous word contained in the same phrase, the specific numerical value which the ambiguous word represents which the attention dialogue person assumes may differ depending on the case.

  In the conventional dialogue support apparatus, as described above, for ambiguous words included in the same phrase, the same specific numerical value is obtained for the interested conversation person, so that the ambiguous word is converted into a specific numerical value. Is difficult to do flexibly.

  The present technology has been made in view of such a situation, and enables an ambiguous word to be flexibly converted, thereby making it possible to easily prevent miscommunication. .

  An information processing apparatus or program according to an aspect of the present technology relates to an ambiguous word detection unit that detects an ambiguous word having an ambiguous degree from input information input from a user, and the ambiguous word from the input information. A task determination unit that determines a task related to the task, an importance determination unit that determines the importance of the task, and the degree of importance represented by the ambiguous word by inputting the importance of the task and the ambiguous word. Learning the classifier that outputs a specific numerical value using the importance, the ambiguous word, and the classifier obtained by using the specific numerical value, the ambiguous word together with the importance An information processing apparatus provided with a conversion unit that converts the ambiguous word into the specific numerical value by giving as an input, or a program for causing a computer to function as the information processing apparatus.

  An information processing method according to one aspect of the present technology detects an ambiguous word having an ambiguous degree from input information input from a user, determines a task related to a destination related to the ambiguous word from the input information, The importance of the task is determined, and learning of a discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word with the importance of the task and the ambiguous word as input is performed. The ambiguous word is converted into the specific numerical value by giving the ambiguous word as an input together with the importance to the discriminator obtained by performing using the word and the specific numerical value. An information processing method including steps.

  In one aspect as described above, an ambiguous word having an ambiguous degree is detected from input information input from a user, and a task related to a destination related to the ambiguous word is determined from the input information. Further, the importance of the task is determined, and the learning of the discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word using the importance of the task and the ambiguous word as input, the importance, By giving the ambiguous word as an input together with the importance to the classifier obtained by performing using the ambiguous word and the specific numerical value, the ambiguous word becomes the specific numerical value. Converted.

  Note that the information processing apparatus may be an independent apparatus or may be an internal block constituting one apparatus.

  The program can be provided by being transmitted via a transmission medium or by being recorded on a recording medium.

  According to one aspect of the present technology, ambiguous words can be flexibly converted.

It is a figure which shows the structural example of one Embodiment of the information processing system to which this invention is applied. It is a figure explaining the outline | summary of a process of information processing system. 2 is a block diagram illustrating a configuration example of a mobile terminal 11. FIG. It is a figure explaining the outline | summary of a process of the portable terminal. It is a figure which shows ambiguous word information. It is a figure which shows task information. It is a figure which shows opponent class information. It is a figure which shows the objective information. It is a figure which shows importance information. It is a figure which shows position information. 4 is a flowchart for explaining processing of the mobile terminal 11. It is a figure which shows learning data. It is a figure which shows the other structural example of the portable terminal. It is a figure explaining the outline | summary of a process of the portable terminal. 4 is a flowchart for explaining processing of the mobile terminal 11. It is a figure which shows learning data. It is a figure which shows the example of the mail which the user A of the portable terminal 11 transmitted (created). It is a figure which shows the example of the mail (blog comment) of the blog which the user A of the portable terminal 11 created, and the mail which the user A transmitted. And FIG. 18 is a block diagram illustrating a configuration example of an embodiment of a computer to which the present technology is applied.

  [One embodiment of an information processing system to which the present technology is applied]

  FIG. 1 shows an information processing system to which the present technology is applied (a system is a logical collection of a plurality of devices, regardless of whether or not each component device is in the same housing). It is a figure which shows the structural example of one Embodiment.

  In FIG. 1, the information processing system includes mobile terminals 11 and 12.

  The mobile terminals 11 and 12 are, for example, mobile phones, smartphones, tablet terminals, and the like.

  In the present embodiment, the mobile terminals 11 and 12 have at least a mail function for exchanging mail (e-mail), for example.

  FIG. 2 is a diagram for explaining the outline of the processing of the information processing system of FIG.

  When the user A of the mobile terminal 11 operates the mobile terminal 11 to create a mail to be transmitted to the user B of the mobile terminal 12, for example, an ambiguous word (ambiguous expression) is included in the mail. Sometimes, in the mobile terminal 11, the ambiguous word included in the mail is converted into a specific numerical value that the user A assumes as a specific numerical value that the ambiguous word represents.

  Then, in the mobile terminal 11, a specific numerical value obtained from the ambiguous word included in the mail (converted from the ambiguous word) is added to the mail (or replaced with the ambiguous word), and the mail is transmitted.

  In FIG. 2, the user A sends a mail (message) including the ambiguous word “a little” “may be a little late” to the user B. At this time, the ambiguous word “a little” is transmitted to the mobile terminal 11. , Converted to “15 minutes”, which is a specific numerical value assumed by user A, and added to the mail.

  Therefore, the user B who has received the mail can recognize that the degree of the ambiguous word “little (delayed)” included in the mail is “15 minutes”, and the user B can recognize the ambiguous word “little”. For example, it is possible to prevent miscommunication caused by assuming a time significantly different from 15 minutes assumed by the user A, such as “5 minutes”.

  [Configuration Example of Mobile Terminal 11]

  FIG. 3 is a block diagram illustrating a configuration example of the mobile terminal 11 of FIG.

  3, the portable terminal 11 includes a dictionary storage unit 21, an ambiguous word DB 22, a task DB 23, an opponent class DB 24, a purpose DB 25, an importance DB 26, a position information DB 27, a control unit 31, a user I / F 32, a mail storage unit 33, Natural language analysis unit 34, ambiguous word detection unit 35, task determination unit 36, opponent class determination unit 37, purpose estimation unit 38, importance determination unit 39, numerical value conversion unit 40, classifier storage unit 41, transmission / reception unit 42, position A detection unit 43, a task achievement determination unit 44, a learning data generation unit 45, a learning data storage unit 46, and a learning unit 47 are included.

  And the control part 31 thru | or the learning part 47 are mutually connected via the bus | bath, and can exchange required data (information) now.

  The dictionary storage unit 21 stores, for example, a dictionary such as a word dictionary or a grammar that is referred to in the natural language analysis performed by the natural language analysis unit 34.

  The ambiguous word DB 22 stores ambiguous word information referred to by ambiguous word detection performed by the ambiguous word detection unit 35.

  The task DB 23 stores task information referred to in task determination performed by the task determination unit 36.

  The partner class DB 24 stores partner class information that is referred to in the partner class determination performed by the partner class determination unit 37.

  The objective DB 25 stores objective information referred to in the objective estimation performed by the objective estimating unit 38.

  The importance DB 26 stores importance information that is referred to in importance determination performed by the importance determination unit 39.

  The position information DB 27 stores position information that is referred to in determining whether the task performed by the task achievement determination unit 44 has been achieved.

  The control unit 31 controls the user I / F 32 or the learning unit 47 as each block connected to the bus.

  The user I / F 32 includes an operation unit (not shown) that can be operated by the user, a display unit (not shown) that displays information to be presented to the user, and the like.

  By operating the user I / F 32 (operation unit thereof), the user can, for example, create a mail or input a command. Further, for example, a mail created by operating the user I / F 32, a mail received by the mobile terminal 11, and other necessary information are displayed on the user I / F 32 (display unit thereof).

  The mail storage unit 33 stores a mail created (and transmitted) by the user operating the user I / F 32 and a mail received by the transmission / reception unit 42. In addition, the mail storage unit 33 stores other information necessary for the mail function (for example, an address book).

  The natural language analysis unit 34 refers to the dictionary stored in the dictionary storage unit 21 and performs natural language analysis such as morphological analysis and syntax analysis on the mail stored in the mail storage unit 33.

  The ambiguous word detection unit 35 refers to the ambiguous word information stored in the ambiguous word DB 22 and the result of the natural language analysis of the mail by the natural language analysis unit 34, and the user stored in the mail storage unit 33 performs the user I / An ambiguous word is detected from the mail created by operating F32.

  The task determination unit 36 is ambiguous that is included in the mail created by the user operating the user I / F 32 based on the task information stored in the task DB 23 and the result of the natural language analysis of the mail by the natural language analysis unit 34. The task related to the destination related to the ambiguous word detected by the word detection unit 35 is determined.

  The partner class determination unit 37 determines the category of the task partner (target) obtained by the task determination unit 36 based on the partner class information stored in the partner class DB 24 and the result of the natural language analysis of the mail by the natural language analysis unit 34. The opponent class that is is determined.

  The purpose estimation unit 38 estimates the purpose of the task obtained by the task determination unit 36 from the purpose information stored in the purpose DB 25 and the result of the natural language analysis of the mail by the natural language analysis unit 34.

  The importance level determination unit 39 refers to the importance level information stored in the importance level DB 26 and the result of the natural language analysis of the mail by the natural language analysis unit 34, and the partner class of the task obtained by the opponent class determination unit 37 Based on the purpose of the task obtained by the purpose estimation unit 38, the importance of the task is determined.

  The numerical value conversion unit 40 inputs the ambiguous word detected by the ambiguous word detection unit 35 together with the importance of the task obtained by the importance determination unit 39 to the classifier stored in the classifier storage unit 41. By giving it, the ambiguous word is converted into a specific numerical value.

  The discriminator storage unit 41 stores a discriminator that receives a task importance level and an ambiguous word and outputs a specific numerical value corresponding to the degree represented by the ambiguous word.

  As the classifier, for example, a multi-class SVM (Support Vector Machine), a neural network, a linear regression model, or any other function can be adopted.

  The transmission / reception unit 42 transmits an email created by the user operating the user I / F 32. Moreover, the transmission / reception part 42 receives the mail addressed to the user A of the mobile terminal 11 transmitted from the mobile terminal 12 (FIG. 1) or the like.

  The position detection unit 43 is composed of, for example, a GPS (Global Positioning System) sensor, and detects the latitude and longitude as the current position of the mobile terminal 11.

  The task achievement determination unit 44 determines whether or not the task obtained by the task determination unit 36 has been achieved (completed) from the position information stored in the position information DB 27 and the current position detected by the position detection unit 43. judge.

  The learning data generation unit 45 generates learning data used for learning (updating) of the discriminator in response to the task achievement determination unit 44 determining that the task has been achieved.

  The learning data storage unit 46 stores the learning data generated by the learning data generation unit 45.

  The learning unit 47 uses the learning data stored in the learning data storage unit 46 to learn (update) the classifier stored in the classifier storage unit 41.

  FIG. 4 is a diagram for explaining the outline of the processing of the mobile terminal 11 of FIG.

  In the mobile terminal 11, the ambiguous word detection unit 35 detects an ambiguous word having an ambiguous degree from, for example, an email created by the user A of the mobile terminal 11 as user input information.

  Further, in the portable terminal 11, the task determination unit 36 determines a task related to the destination related to the ambiguous word from, for example, mail stored in the mail storage unit 33 as user input information, and determines the importance level. In part 39, the importance of the task is determined.

  And in the portable terminal 11, in the numerical value conversion part 40, an ambiguous word is converted into a specific numerical value by giving an ambiguous word with an importance as an input with respect to the discriminator memorize | stored in the discriminator memory | storage part 41. Is done.

  For example, it is assumed that mails M # 1 and M # 2 as shown in FIG. 4 are exchanged between the user A of the mobile terminal 11 and the user B of the mobile terminal 12.

  The mail M # 1 is a mail transmitted from the user A (the mobile terminal 11) on January 11, 2011 and received by the user B (the mobile terminal 12). ) At 2 o'clock, there is a message that suggests a meeting in front of Alta in Shinjuku “Nja Tomorrow is a meeting in front of Alta at 2 o'clock!”.

  The mail M # 2 is a reply mail sent from the user B and received by the user A on January 11, 2011, to the mail M # 1, and a message “acknowledging the proposal for waiting by the mail M # 1”. Ryokai "is described.

  E-mail M # 3 is an e-mail sent from User A and received by User B at 13:45 on January 12, 2011, which is the day of the meeting, and is notified that it will be delayed after the meeting time (2 o'clock) The message “It may be a little late (about 9 minutes ± 3 minutes)” is described.

  That is, for example, user A is in Shibuya around 13:45 on January 12, 2011, the day of the meeting, and will be in time for 14:00 (2 pm), the meeting time at Alta in Shinjuku. Therefore, it is assumed that the user A operates the user I / F 32 of the mobile terminal 11 and creates the mail M # 3 with the message “Slightly delayed” as the text for the user B.

  In this case, in the mobile terminal 11, the ambiguous word detection unit 35 detects the ambiguous word “a little” from the mail M # 3 created by the user A (its body “a little delayed”).

  Further, in the mobile terminal 11, the task determination unit 36 determines that the mail M # 3 and, if necessary, the related mail that is mail related to the mail M # 3, that is, the destination of the mail M # 3, for example. From the emails M # 1 and M # 2 exchanged with user B, the task related to the relationship "delayed mom" with the ambiguous word "a little" is a waiting task (a waiting task) It is determined that

  Further, the importance level determination unit 39 determines the importance level of the waiting task determined from the mail M # 3 or the like.

  In the mobile terminal 11, the ambiguous word “little” is added to the classifier stored in the classifier storage unit 41 in the numerical value conversion unit 40 together with the importance level of the waiting task obtained by the importance level determination unit 39. By giving it as an input, the ambiguous word “slightly” representing the degree of time delayed from the meeting time is converted to a specific numerical value “9 minutes”.

  Further, in the control unit 31, a specific numerical value “9 minutes” obtained by converting the ambiguous word “little” is added to the mail M # 3 created by the user A, and the specific numerical value “9 minutes” is added. M # 3 is transmitted from the transmission / reception unit 42.

  Here, in FIG. 4, a specific numerical value “9 minutes” obtained by converting the ambiguous word “little” into the mail M # 3, that is, a specific time “9 minutes” delayed from the waiting time, and the margin of the time. The margin time “± 3 minutes” is also described.

  As described above, in the mobile terminal 11, a specific numerical value “9 minutes” obtained by converting the ambiguous word “little” is added to the mail M # 3 created by the user A, so that the user who viewed the mail M # 3 B can recognize that the ambiguous word “slightly” indicating the degree of time delayed from the meeting time is specifically about 9 minutes, and miscommunication occurs between users A and B. This can be prevented.

  Note that the mobile terminal 12 of the user B that receives the mail M # 3 from the user A only needs to have a general mail function, and does not need to be configured in the same manner as the mobile terminal 11 (however, It may be configured similarly to the portable terminal 11).

  FIG. 5 is a diagram showing an example of ambiguous word information stored in the ambiguous word DB 22 of FIG.

  In the ambiguous word information, a word (word notation) as an ambiguous word is associated with a semantic class of the ambiguous word and a dependency destination keyword.

  In the ambiguous word information, the meaning class of the ambiguous word is a class for classifying the meaning of the ambiguous word, and the related keyword of the ambiguous word is a word that can be related to the ambiguous word.

  In the ambiguous word detection unit 35, for example, using the result of the natural language analysis of the mail by the natural language analysis unit 34, the notation and meaning are the word notation and meaning of the ambiguous word information from the words constituting the mail. Words that match the class are detected as ambiguous word candidates.

  When the ambiguous word detection unit 35 can detect the ambiguous word candidate from the mail, the ambiguous word candidate has ambiguous word information (hereinafter also referred to as “match ambiguous word information”) having a word notation and a semantic class that the ambiguous word candidate matches. It is determined whether or not the relationship destination keyword of the parenthesis exists in the mail as a relationship destination of an ambiguous word candidate.

  When the destination keyword included in the match ambiguous word information does not exist as the destination of the ambiguous word candidate in the mail, the ambiguous word detection unit 35 recognizes that the mail has no ambiguous word.

  Further, when the destination keyword included in the match ambiguous word information exists as a destination of the ambiguous word candidate in the mail, the ambiguous word detection unit 35 detects the ambiguous word candidate as an ambiguous word included in the mail ( recognize.

  For example, when the message “may be slightly delayed” is described in the mail, the ambiguous word detection unit 35 detects “little” from the mail as a candidate for the ambiguous word.

  In an email in which the message “A little late” is described, the candidate for the ambiguous word candidate “a little” is the word “late”, and the word “a late” is the destination “a little” in FIG. Present in the keyword.

  In this case, the ambiguous word detection unit 35 detects “little” from the mail as an ambiguous word.

  FIG. 6 is a diagram showing an example of task information stored in the task DB 23 of FIG.

  In the task information, a keyword is associated with a condition category.

  In the task information, the keyword is a word used to determine a task related to the relation destination related to the ambiguous word detected by the ambiguous word detection unit 35, and the condition category represents an attribute of the keyword.

  In the task determination unit 36, a keyword included in the task information from an email (hereinafter also referred to as a detection email) in which an ambiguous word is detected by the ambiguous word detection unit 35, and further, a related email related to the detection email, if necessary. And the task related to the relation related to the ambiguous word is determined based on the condition category of the keyword.

  That is, the task determination unit 36 has a template for a condition category for each task.

  For example, a template “date / time: location: purpose:” is prepared for the waiting task, and the task determination unit 36 uses a keyword of a condition category representing the date / time (waiting) from the detection mail and the related mail, ( Attempts to extract a keyword of a condition category representing a place (meeting) and a keyword of a condition category representing a purpose.

  For example, if the detected mail is the mail M # 3 in FIG. 4, the task determination unit 36 represents the date and time from the mail M # 3 and the related mails M # 1 and M # 2. The date and time of meeting “(2011) 12:00 on January 12 (2 pm)” is extracted as the keyword for the condition category, and the keyword “ No) Alta ”is extracted.

  When the task determination unit 36 extracts a keyword, the task determination unit 36 embeds the keyword in a template. In this case, the keyword “date and time: place: purpose:” is embedded in the template, so that the template “date and time: 12:00 on January 12th, place:“ Shinjuku Alta ”, purpose: nothing in particular” is obtained.

  If a keyword is embedded in at least “date and time” and “location” for the template “date and time: location: purpose:” of the waiting task, the task determination unit 36 determines that the ambiguous word detected from the detected mail is the relevant destination. It is determined that the related task is a waiting task.

  FIG. 7 is a diagram illustrating an example of partner class information stored in the partner class DB 24.

  In the partner class information, the person ID and the partner class are associated with each other.

  In the partner class information, the person ID is information for specifying a person (here, the partner who is the destination of the mail), and for example, the name of the person can be adopted. The opponent class represents the category (class) of the person specified by the person ID.

  The partner class information can be manually input by the user, for example.

  Specifically, for example, a name registered in the address book of the mail function may be adopted as a person ID, and the person's partner class specified by each person ID may be manually input by the user. it can.

  Here, the person specified by the person ID can be clustered on the basis of, for example, the usage rate of honorifics or the frequency (familiarity) of email exchanges as information on the style of email addressed to the person. It is possible to adopt the clustering result of the clustering as a partner class.

  Also, the partner class manually input by the user can be corrected according to the clustering result by the clustering as described above.

  In the partner class determination unit 37, in the partner class information, for example, the partner class (for example, “(close) friend”) associated with the person ID of the partner who is the destination of the detected mail is the task determination unit. It is determined that it is the opponent class of the opponent (target) of the task obtained in 36.

  In the portable terminal 11, the partner class is determined based on a hypothesis that the weight of the word will change depending on the partner who is the destination of the detected mail.

  That is, for example, even if the same message is “elongate a little longer”, the e-mail to the manager of the company and the e-mail to a close friend are specific to the extent that the ambiguous word “little” is included in the e-mail. Expected to be different. Therefore, in the portable terminal 11, even if the phrase is the same, the partner class is determined so that a different numerical value can be obtained as a specific numerical value represented by the ambiguous word included in the phrase depending on the partner class. The

  FIG. 8 is a diagram illustrating an example of purpose information stored in the purpose DB 25.

  In the purpose information, a keyword is associated with a target category (purpose category) represented by the keyword.

  In the purpose estimating unit 38, for example, a keyword included in the purpose information is extracted from the detection mail and the related mail, and each time the keyword is extracted, the appearance frequency of the target category associated with the keyword is counted up. Is done.

  Then, the purpose estimation unit 38 estimates, for example, the purpose category having the highest appearance frequency as the purpose of the task obtained by the task determination unit 36 based on the appearance frequency of the purpose category.

  When the appearance frequency of all the target categories is 0, it is estimated that there is no purpose of the task (“purpose: none in particular”).

  In addition to the method of counting the appearance frequency of the target category associated with the keyword as described above, the task purpose estimation method includes mining a large amount of emails and using machine learning techniques such as SVM. It is possible to adopt a method of generating a discriminator by using them and performing clustering.

  Here, in the portable terminal 11, the purpose of the task is estimated based on a hypothesis that the degree (scale) represented by the ambiguous word will vary depending on the purpose of the task.

  That is, for example, even if the same message is “slightly delayed” addressed to the same friend, an email for shopping purposes and an email for movies and concerts with a fixed time Then, it is expected that the specific numerical value to the extent that the ambiguous word “a little” contained in the email represents is different. Therefore, in the portable terminal 11, even if the phrase is the same, depending on the purpose of the task, the purpose of the task is such that different numerical values can be obtained as specific numerical values represented by the ambiguous words included in the phrase. Presumed.

  FIG. 9 is a diagram illustrating an example of importance information stored in the importance DB 26.

  In the importance level information, the category and the importance level of the category are associated with each other.

  Here, the category included in the importance level information includes at least the purpose of the task and the partner class of the task.

  Further, in the importance level information of FIG. 9, the category includes a keyword.

  In the importance level determination unit 39, in the importance level information, the importance level (hereinafter also referred to as the importance level of the opponent class) associated with the category that matches the opponent class obtained in the opponent class determination unit 37, and the purpose The importance level (hereinafter also referred to as the target importance level) associated with the category that matches the purpose obtained by the estimation unit 38 is detected.

  Then, the importance level determination unit 39 adds the importance level of the opponent class and the target importance level, and determines the added value obtained as a result as the task importance level obtained by the task determination unit 36.

  In addition, in the importance level determination unit 39, for example, each of the importance level of the opponent class and the target importance level is multiplied by a coefficient (weight), and linear combination is performed. The importance can be determined.

  Further, the importance level determination unit 39 can extract keywords as categories included in the importance level information from the detection mail and the related mail, and can extract the keywords from the detection mail and the related mail in the importance level information. The importance of the task can be determined by using the importance associated with the keyword as the category in the same manner as the importance of the opponent class and the importance of the target class.

  For example, in the case of a waiting task, the importance of the task corresponds to a level at which the user is conscious that the user should not be late for the waiting time.

  FIG. 10 is a diagram illustrating an example of position information stored in the position information DB 27.

  In the position information, a keyword representing a place is associated with the latitude and longitude of the place represented by the keyword.

  For example, the task achievement determination unit 44 recognizes the latitude and longitude of the meeting place by referring to the position information.

  Then, the task achievement determination unit 44 obtains the distance between the current position and the meeting place based on the latitude and longitude as the current position detected by the position detection unit 43 and the latitude and longitude of the meeting place, If the distance is within a predetermined threshold TH meter, it is determined that the waiting task has been achieved (completed).

  [Processing of mobile terminal 11]

  FIG. 11 is a flowchart for explaining processing of the mobile terminal 11 of FIG.

  In step S <b> 11, the control unit 31 determines whether or not a mail is input from the user A, that is, the user A operates the user I / F 32 to create a mail and further transmit the mail. It is determined whether the user I / F 32 has been operated.

  If it is determined in step S11 that user A has not input a mail, the process returns to step S11.

  If it is determined in step S11 that the mail is input from the user A, the control unit 31 stores the mail from the user A in the mail storage unit 33, and the process proceeds to step S12. .

  Here, in addition to the mail input from the user A, the mail storage unit 33 stores the mail addressed to the user A received by the transmission / reception unit 42.

  In step S12, the natural language analysis unit 34 performs the natural language analysis of the noticed mail as the noticed noticed mail, which is the mail input from the user A stored in the mail storage unit 33 immediately before. The process proceeds to step S13.

  In step S13, the ambiguous word detection unit 35 refers to the ambiguous word information (FIG. 5) stored in the ambiguous word DB 22 and the result of the natural language analysis of the attention mail by the natural language analysis unit 34, so that the attention mail is extracted. The ambiguous word is detected, and the process proceeds to step S14.

  In step S <b> 14, the control unit 31 determines whether or not an ambiguous word is detected from the attention mail in the ambiguous word detection unit 35.

  If it is determined in step S14 that an ambiguous word has not been detected from the noticed mail, that is, the ambiguous word is not included in the noticed mail, or an ambiguous word is included. If there is no relationship, the process proceeds to step S15, the transmission / reception unit 42 transmits the attention mail (wirelessly), and the process returns to step S11.

  Therefore, when an ambiguous word is not detected from the attention mail, the mail (attention mail) input by the user A is transmitted as it is.

  On the other hand, if it is determined in step S14 that an ambiguous word has been detected from the noticed email, the process proceeds to step S16, and the natural language analysis unit 34 of the emails stored in the email storage unit 33, Performs natural language analysis of related emails related to the email of interest.

  In other words, the natural language analysis unit 34 extracts, from the mail stored in the mail storage unit 33, the mail whose destination or sender matches the destination of the noticed mail, the mail whose subject includes the subject of the noticed mail, or the like. Select related emails related to.

  Then, the natural language analysis unit 34 performs natural language analysis of the related mail, and the process proceeds from step S16 to step S17.

  In step S17, the task determination unit 36 refers to the task information (FIG. 6) stored in the task DB 23, the attention mail by the natural language analysis unit 34, and the result of the natural language analysis of the related mail, thereby paying attention. The task related to the destination of the ambiguous word of the mail (for example, “slightly” of the mail M # 3 in FIG. 4) (for example, “delayed” in the mail M # 3 of FIG. 4) is determined. Proceed to

  In step S <b> 18, the control unit 31 determines whether the task determination unit 36 has been able to determine a task (hereinafter also referred to as a related task) related to the destination of the ambiguous word of the target email.

  If it is determined in step S18 that the related task could not be determined, the process proceeds to step S15, the transmission / reception unit 42 transmits the attention mail, and the process returns to step S11.

  Therefore, when the related task cannot be determined, the mail (attention mail) input by the user A is transmitted as it is.

  On the other hand, if it is determined in step S18 that the related task has been determined, the process proceeds to step S19, where the opponent class determination unit 37 stores the opponent class information (FIG. 7) stored in the opponent class DB 24, etc. The partner class of the related task is determined by referring to, and the process proceeds to step S20.

  Here, in order to simplify the description, it is assumed that the related task is a waiting task.

  In step S20, the purpose estimation unit 38 refers to the purpose information (FIG. 8) stored in the purpose DB 25, the attention mail by the natural language analysis unit 34, and the result of the natural language analysis of the related mail, thereby The purpose of the waiting task, which is a task, is estimated, and the process proceeds to step S21.

  In step S21, the importance level determination unit 39 refers to the importance level information (figure) stored in the importance level DB 26 and the result of the natural language analysis of the mail by the natural language analysis unit 34, thereby waiting for a related task. The importance of the waiting task is determined based on the task partner class, purpose, etc., and the process proceeds to step S22.

  In step S22, the control unit 31 determines whether the classifier stored in the classifier storage unit 41 has already been learned.

  If it is determined in step S22 that the classifier stored in the classifier storage unit 41 has already been learned, the process proceeds to step S23, and the numerical value conversion unit 40 stores the classifier in the classifier storage unit 41. The ambiguous word detected by the ambiguous word detection unit 35 using the identified classifier, the ambiguous word detected by the ambiguous word detection unit 35, and the importance of the waiting task obtained by the importance level determination unit 39, Convert to a specific number.

  That is, in step S23, the numerical value conversion unit 40 waits for the classifier stored in the classifier storage unit 41 with the ambiguous word detected by the ambiguous word detection unit 35 obtained by the importance level determination unit 39. By giving it as an input together with the importance of the task, a specific numerical value obtained by converting the ambiguous word is obtained as an output of the discriminator.

  And the control part 31 reflects the specific numerical value obtained in the numerical value conversion part 40 by adding to the attention mail memorize | stored in the mail memory | storage part 33, and a process progresses to step S24 from step S23.

  In step S24, the transmission / reception unit 42 transmits the attention mail, and the process proceeds to step S25.

  In this case, the mail (attention mail) input by the user A is transmitted with a specific numerical value obtained by converting the ambiguous word included in the mail.

  On the other hand, if it is determined in step S22 that the classifier stored in the classifier storage unit 41 has not been learned, the process skips step S23 and proceeds to step S24. The process proceeds to step S25.

  In this case, the mail (attention mail) input by the user A is transmitted as it is.

  In step S25, the task achievement determination unit 44 is obtained by the task determination unit 36 by referring to the position information (FIG. 10) stored in the position information DB 27 and the current position detected by the position detection unit 43. Determine whether the waiting task has been accomplished.

  If it is determined in step S25 that the waiting task has not been achieved, for example, the process waits for a few seconds, and the process returns to step S25.

  If it is determined in step S25 that the waiting task has been achieved, that is, for example, the distance between the waiting location in the waiting task and the current position is within a predetermined threshold TH meter, and user A If it can be considered that the place has arrived, the process proceeds to step S26, and the learning data generation unit 45 generates learning data used for learning (updating) of the classifier stored in the classifier storage unit 41.

  In other words, the learning data generation unit 45 subtracts the waiting time in the waiting task from the time at which the waiting task is achieved, so that the user A arrives at the waiting place (the delay time actually delayed) Hereinafter, it is also referred to as an actual delay time).

  Specifically, for example, the waiting time (waiting time) is 14:00 (00:00), and the time when the user A arrives at the meeting place (waiting place) is 14:11:15. If there is, the actual delay time is 11 minutes and 15 seconds (= 14: 11: 15-14: 14).

  Then, the learning data generation unit 45 stores the ambiguous word detected from the attention mail, the importance of the waiting task, and the actual delay time in the learning data storage unit 46 as learning data. Proceed to step S27.

  In step S27, the learning unit 47 learns (updates) the classifier stored in the classifier storage unit 41 using the learning data stored in the learning data storage unit 46, and the process returns to step S11. .

  That is, the learning unit 47 gives the ambiguous word as the learning data, the importance of the waiting task, and the importance of the waiting task as the input of the discriminator as well as the actual delay time. The delay time is given as an output of the discriminator, and the discriminator is learned.

  As described above, for example, a multi-class SVM, a neural network, or a linear regression model can be adopted as the discriminator.

  When a linear regression model is adopted as a discriminator, for example, a linear regression model is prepared for each ambiguous word, and an actual delay time (a value close to) is obtained as an output for the input of the importance level of the waiting task. The coefficient of the linear regression model that can be obtained (the coefficient of the linear linear expression that performs the product-sum operation of the importance of the waiting task and the coefficient) is obtained.

  In this case, in the numerical conversion unit 40, in step S23, a linear regression model for the ambiguous word detected from the noticed mail is selected, and the importance of the queuing task is given as an input to the linear regression model. (A linear linear expression) is obtained as a specific numerical value for the ambiguous word detected from the notice mail.

  In addition, when multi-class SVM or neural network is used as a classifier, ambiguous words and importance of waiting task are used as input of multi-class SVM or neural network, and output of multi-class SVM or neural network. As described above, machine learning is performed using the actual delay time.

  In this case, in the numerical value conversion unit 40, in step S23, the ambiguous word detected from the noticed mail and the importance of the waiting task are given as inputs of the multiclass SVM or neural network, and the multiclass SVM or the The output of the neural network is obtained as a specific numerical value for the ambiguous word detected from the notice mail.

  As an input to the classifier, in addition to the ambiguous word and the importance of the waiting task, for example, in the ambiguous word information stored in the ambiguous word DB 22 (FIG. 5), the semantic class associated with the ambiguous word, etc. Can be used.

  FIG. 12 is a diagram illustrating an example of learning data generated by the learning data generation unit 45 and stored in the learning data storage unit 46.

  The learning data generation unit 45 generates learning data including the ambiguous word detected from the noticed mail, the importance of the waiting task, and the actual delay time every time the waiting task is achieved. Remembered.

  Then, by repeating the learning (updating) of the discriminator using such learning data, a discriminator that outputs robust specific values that are robust with respect to the ambiguous word used by the user A is configured. It will be done.

  As a discriminator, for example, when a linear regression model is adopted, the maximum value or minimum value of the difference (error) between the output of the linear regression model and the actual delay time given as learning data is obtained, and The maximum value and the minimum value can be used as the margin time described with reference to FIG.

  For example, when a multiclass SVM is adopted as the discriminator, the upper limit value and the lower limit value of the class of the multiclass SVM can be used as the margin time described in FIG.

  As described above, in the mobile terminal 11, an ambiguous word (for example, “a little”) is detected from the attention mail as the input information input from the user, and further, the attention mail as the input information input from the user From the related mail, a related task (for example, a waiting task) related to the destination related to the ambiguous word is determined.

  Further, in the portable terminal 11, the importance level of the waiting task is determined based on the partner class and purpose of the waiting task, and corresponds to the degree represented by the ambiguous word “a little” according to the completion (achievement) of the waiting task. The actual delay time as a specific numerical value is obtained (generated).

  The mobile terminal 11 learns the classifier using the importance of the waiting task, the ambiguous word, and the actual delay time as a specific numerical value.

  Therefore, in the discriminator, for the ambiguous word, the time (actual delay time) in which the user is actually delayed is acquired according to the importance of the waiting task, that is, in the case of the partner or purpose of the waiting.

  Further, the portable terminal 11 obtains a value obtained by converting the ambiguous word into a specific numerical value by giving the ambiguous word as an input together with the importance of the waiting task to such a discriminator.

  Therefore, even if it is an ambiguous word included in the same phrase of the user A, in some cases, for example, a meeting partner (for example, a work partner or a friend), a purpose of meeting ( The ambiguous word can be flexibly converted into a specific numerical value of a different value depending on whether or not the meeting is such that there is no problem even if it is delayed after the meeting time, and miscommunication can be prevented.

  Moreover, in the conventional dialogue support apparatus described in Patent Document 1, it is necessary to manually create a conversion table in which an ambiguous word is associated with a range of specific numerical values to the extent that the ambiguous word represents. In the mobile terminal 11, the classifier is learned using the importance of the waiting task, the ambiguous word, and the actual delay time as a specific numerical value. Therefore, like a conventional dialogue support device, There is no need to create a translation table. Therefore, the user can easily prevent miscommunication.

  Furthermore, as learning of the classifier progresses (by learning of the classifier using a large number of learning data), in the classifier, various related tasks can be compared with the partner class and purpose of the related task. Since the specific numerical value to the extent represented by the ambiguous word used by the user is obtained, the accuracy of conversion of the ambiguous word into the specific numerical value by the classifier can be improved. That is, an ambiguous word used by a user can be accurately converted into a specific numerical value assumed by the user.

  The classifier learns (updates) each time task importance, ambiguous words, and specific numerical values are obtained, so that the specific numerical values that the user assumes for a certain ambiguous word are Even if it changes according to the course (change in age), it is possible to convert the ambiguous word into a specific numerical value following the change.

  In addition, by using a discriminator, for example, messages that contain ambiguous words related to actions (for example, "wait" or "end"), such as "wait for a moment" or "finish a little later" In the manual work, it is possible to convert an ambiguous word that is difficult to input accurately from a specific numerical value to the extent that the ambiguous word represents into a relatively accurate specific numerical value.

  In the above case, the importance level of the waiting task is determined based on both the partner class and purpose of the waiting task. However, the importance level of the waiting task is determined only by the partner class of the waiting task. It is possible to obtain it based only on the purpose.

  Further, the mobile terminal 12 is configured in the same manner as the mobile terminal 11, and an identifier for converting an ambiguous word used by the user A obtained by learning in the mobile terminal 11 into a specific numerical value (user A's identifier). Can be provided (transmitted) to the mobile terminal 12 of the user B, and the mobile terminal 12 can convert the ambiguous word included in the mail from the user A into a specific numerical value using the identifier of the user A. It is.

  Moreover, in the portable terminal 11, the discriminator can be prepared for each related task or can be prepared for each of a plurality of related tasks.

  Further, the classifier can store a standard classifier, which is a standard classifier, in the classifier storage unit 41, and can update the standard classifier using the learning data.

  The standard classifier can be generated, for example, by collecting learning data from a plurality of users on a server on a network such as the Internet and learning the classifier using the learning data.

  Since the classifier is not learned immediately after the user A purchases the portable terminal 11, the standard classifier is generated in the server on the network as described above, and the standard is identified in the portable terminal 11. By downloading the discriminator, it is possible to realize a hot start that enables conversion of ambiguous words immediately after purchase of the mobile terminal 11.

  In the server on the network, users can be classified into several categories according to the user's profile, and a standard classifier can be generated for each category using learning data from users belonging to that category.

  In this case, the server on the network can provide the mobile terminal 11 with a standard identifier of the category of the user A obtained from the profile of the user A of the mobile terminal 11.

  [Another configuration example of the mobile terminal 11]

  FIG. 13 is a block diagram showing another configuration example of the mobile terminal 11 of FIG.

  In FIG. 13, portions corresponding to those in FIG. 3 are denoted by the same reference numerals, and description thereof will be omitted below as appropriate.

  The mobile terminal 11 in FIG. 13 includes the dictionary storage unit 21 to the position information DB 27 and the control unit 31 to the learning unit 47, and is common to the case of FIG. 13, and includes an acceleration center 61, a behavior analysis unit 62, a behavior log. It differs from the case of FIG. 13 in having a storage unit 63 and a numerical value correction unit 64.

  The acceleration sensor 61 detects the acceleration of the mobile terminal 11 (the user A who owns it).

  The behavior analysis unit 62 performs a behavior analysis of the user A based on the current location of the user A detected by the position detection unit 43 and the acceleration of the user A detected by the acceleration sensor 61, and as a result of the behavior analysis, Action information indicating what action the user is taking is obtained.

  Here, the action information is information such as that the user A is walking, running, riding a train, riding a car, and the like.

  The behavior log storage unit 63 stores a log of behavior information (behavior log) obtained by the behavior analysis unit 62.

  The numerical value correction unit 64 performs numerical value correction using a specific numerical value obtained by converting the ambiguous word in the numerical value conversion unit 40.

  FIG. 14 is a diagram for explaining the outline of the processing of the mobile terminal 11 of FIG.

  Now, the user A of the mobile terminal 11 is meeting with the user B of the mobile terminal 12 at 14:00 in Shinjuku Alta, and the user A is traveling to Shinjuku (the Alta) where the meeting is held by train. Suppose that

  In addition, it is assumed that the user A sends a message “slightly delayed” to the user B by e-mail at 13:55 on the train that departs from Shibuya Station.

  In this case, in the mobile terminal 11 of FIG. 13, the mail transmitted from the user A to the user B is processed as the notice mail.

  That is, in the mobile terminal 11, the behavior analysis unit 62 indicates that the user A is on the train based on the user's acceleration detected by the acceleration sensor 61, that is, the (current) moving means of the user A It is recognized that.

  Further, in the mobile terminal 11, the behavior analysis unit 62 determines that the current position detected by the position detection unit 43, that the moving means of the user A is a train, and a meeting place recognized from the attention mail and the related mail. Based on Shinjuku (Alta), the estimated required time required to arrive at the meeting place is obtained.

  Here, the estimated required time when the user A is moving by train is obtained by, for example, inquiring a transfer guide API (Application Program Interface) that provides a transfer guide provided by a server on the Internet. Can be sought.

  In FIG. 14, 13 minutes is obtained as the estimated required time.

  The behavior analysis unit 62 calculates the predicted arrival time predicted to arrive at the meeting place by adding the estimated required time to the current time.

  In FIG. 14, 14:08 is obtained as the predicted arrival time by adding 13 minutes that is the estimated required time to 13:55 that is the current time.

  The behavior analysis unit 62 subtracts the waiting time from the predicted arrival time to obtain a predicted delay time that is a predicted value of the delay time delayed from the waiting time.

  In FIG. 14, 8 minutes is obtained as the predicted delay time by subtracting the waiting time of 14:00 from the predicted arrival time of 14:08.

  Thereafter, it is assumed that the user A actually arrives at the meeting place at 14:05.

  Whether or not the user A has arrived at the meeting place depends on whether or not the distance between the meeting place and the current position is within a predetermined threshold TH meter in the task achievement determination unit 44 as described in FIG. Determined.

  Here, the time when the user A actually arrives at the meeting place is also referred to as the actual arrival time.

  The behavior analysis unit 62 subtracts the waiting time from the actual arrival time to obtain an actual delay time that is actually delayed from the waiting time.

  In FIG. 14, 5 minutes is obtained as the actual delay time by subtracting the waiting time of 14:00 from the actual arrival time of 14:05.

  The behavior analysis unit 62 obtains a correction value (specific numerical value) for correcting the predicted delay time by subtracting the predicted delay time from the actual delay time.

  In FIG. 14, by subtracting 8 minutes which is the predicted delay time from 5 minutes which is the actual delay time, −3 minutes is obtained as a correction value for correcting the predicted delay time.

  Further, in the portable terminal 11 of FIG. 13, as in the case of FIG. 3, the ambiguous word “little” is detected from the 13:55 mail “slightly delayed” sent by the user A to the user B, It is required that the related task is a waiting task and the importance of the waiting task.

  In the mobile terminal 11 of FIG. 13, when the moving means of the user A, the correction value for correcting the predicted delay time, the ambiguous word “little”, and the importance of the waiting task are obtained as described above, the learning unit 47 In (1), learning (updating) of the classifier is performed.

  3, the ambiguous word and the importance of the waiting task are input to the discriminator, and the actual delay time as a specific numerical value is used as the output of the discriminator to learn the discriminator. However, in the mobile terminal of FIG. 13, the ambiguous word, the importance of the queuing task, and the moving means of the user A are input to the discriminator, and a correction value as a specific numerical value is As an output, learning (updating) of the discriminator stored in the discriminator storage unit 41 is performed.

  Then, after learning of the classifier is performed in the mobile terminal 11 of FIG. 13, when the waiting as described in FIG. 14 is performed again between the users A and B, the same as described above. Thus, the moving means of the user A, the predicted delay time, the ambiguous word “little”, and the importance of the waiting task are obtained.

  Then, the numerical value conversion unit 40 gives the discriminator stored in the discriminator storage unit 41 by inputting the ambiguous word, the importance level of the waiting task, and the moving means of the user A as input. A correction value for correcting the predicted delay time to be output is obtained.

  Then, the numerical value correction unit 64 corrects the predicted delay time by adding the correction value obtained by the numerical value conversion unit 40 to the predicted delay time, and the control unit 31 corrects the predicted delay time after correction (hereinafter, referred to as “corrected predicted delay time”). Correction delay time) is added to the mail as a specific numerical value obtained by converting the ambiguous word.

  [Processing of mobile terminal 11]

  FIG. 15 is a flowchart for explaining processing of the mobile terminal 11 of FIG.

  In FIG. 15, in steps S51 to S61, the same processing as in steps S11 to S21 of FIG. 11 is performed, whereby the ambiguous word included in the noticed mail, the related task is a waiting task, and the waiting task. And the process proceeds to step S62.

  In step S <b> 62, the behavior analysis unit 62 performs a behavior analysis of the user A, thereby obtaining a moving means for the user A. Further, the behavior analysis unit 62 obtains a predicted delay time as described with reference to FIG. 14, and the process proceeds from step S62 to step S63.

  In step S63, the control unit 31 determines whether or not the classifier stored in the classifier storage unit 41 has already been learned.

  If it is determined in step S63 that the classifier stored in the classifier storage unit 41 has already been learned, the process proceeds to step S64, and the numerical value conversion unit 40 stores the classifier in the classifier storage unit 41. The ambiguous word is converted into a specific value of a correction value for correcting the predicted delay time by using the identified classifier, the ambiguous word included in the notice mail, the importance of the waiting task, and the moving means of the user A.

  That is, in step S64, the numerical value conversion unit 40 sends the ambiguous word included in the noticed email to the classifier stored in the classifier storage unit 41 together with the importance of the waiting task and the moving means of the user A. By giving it as an input, a specific numerical value obtained by converting the ambiguous word included in the notice mail into a correction value is obtained as an output of the discriminator.

  Thereafter, the process proceeds from step S64 to step S65, and the numerical value correction unit 64 corrects the predicted delay time obtained by the behavior analysis unit 62 with the correction value obtained by the numerical value conversion unit 40. Then, the control unit 31 adds the correction delay time (specific numerical value) obtained by the numerical value correction unit 64 correcting the predicted delay time with the correction value to the notice mail stored in the mail storage unit 33. The process proceeds from step S65 to step S66.

  In step S66, the transmission / reception unit 42 transmits the attention mail, and the process proceeds to step S67.

  In this case, the mail (attention mail) input by the user A is transmitted with a specific numerical value obtained by converting the ambiguous word included in the mail into the corrected delay time.

  On the other hand, if it is determined in step S63 that the discriminator stored in the discriminator storage unit 41 has not been learned, the process skips steps S64 and S65 and proceeds to step S66. The attention mail is transmitted, and the process proceeds to step S67.

  In this case, the mail (attention mail) input by the user A is transmitted as it is.

  In step S67, the task achievement determination unit 44 refers to the position information (FIG. 10) stored in the position information DB 27 and the current position detected by the position detection unit 43, as in step S25 of FIG. Thus, it is determined whether or not the waiting task obtained by the task determination unit 36 has been achieved.

  If it is determined in step S67 that the waiting task has not been achieved, for example, the process waits for a few seconds, and the process returns to step S67.

  If it is determined in step S67 that the waiting task has been achieved, that is, for example, the distance between the waiting location in the waiting task and the current position is within a predetermined threshold TH meter, and user A If it can be considered that the place has arrived, the process proceeds to step S68, and the learning data generation unit 45 generates learning data used for learning (updating) of the classifier stored in the classifier storage unit 41.

  In other words, the learning data generation unit 45 subtracts the waiting time in the waiting task from the time at which the waiting task is achieved, so that the user A arrives at the waiting place (the delay time actually delayed) (Actual delay time).

  Further, the learning data generation unit 45 obtains a correction value for correcting the predicted delay time by subtracting the predicted delay time obtained by the behavior analysis unit 62 from the actual delay time.

  Then, the learning data generation unit 45 stores the ambiguous word included in the attention email, the importance of the waiting task, the moving means of the user A, and the correction value in the learning data storage unit 46 as learning data. The process proceeds from step S68 to step S69.

  In step S69, the learning unit 47 learns (updates) the classifier stored in the classifier storage unit 41 using the learning data stored in the learning data storage unit 46, and the process returns to step S51. .

  That is, the learning unit 47 uses the ambiguous word as the learning data, the importance of the waiting task, the moving means of the user A, and the ambiguous word, the importance of the waiting task among the correction values, and the moving means of the user A. Is given as an input of the discriminator and a correction value is given as an output of the discriminator to perform learning of the discriminator.

  FIG. 16 is a diagram illustrating an example of learning data generated by the learning data generation unit 45 of FIG. 13 and stored in the learning data storage unit 46.

  Each time the waiting task is achieved, the learning data generation unit 45 includes ambiguous words included in the noticed email, the importance of the waiting task, the moving means (of user A), and a correction value for correcting the predicted delay time. Learning data is generated and stored in the learning data storage unit 46.

  Then, by repeating the learning (updating) of the discriminator using such learning data, a discriminator that outputs robust specific values that are robust with respect to the ambiguous word used by the user A is configured. It will be done.

  As described above, in the mobile terminal 11 of FIG. 13, the moving means of the user A is given as the input of the discriminator, and the correction value is obtained for the predicted delay time when the moving means is used. Regardless of the moving means, the ambiguous word can be accurately converted into a specific numerical value assumed by the user.

  In the above, mail is used as the input information from the user. However, as the input information from the user, for example, the voice recognition result obtained by voice recognition of the user's dialogue or the user's use is used in addition to the mail. Electronic schedule books, user blogs, etc. can be employed.

  In addition, the related task is not limited to the waiting (collection) task.

  Here, in FIG. 3 (also in FIG. 13), the dictionary storage unit 21 is a database with relatively high access frequency, and the discriminator storage unit 41 and the learning data storage unit 46 are updated with relatively high frequency. It is a dynamic database. Further, the opponent class DB 24 is an individual database for each user.

  On the other hand, the ambiguous word DB 22, the task DB 23, the purpose DB 25, the importance DB 26, and the position information DB 27 are so-called static databases that are common to users and have a relatively low update frequency.

  The ambiguous word DB 22, task DB 23, purpose DB 25, importance DB 26, and position information DB 27, which are static databases, can be provided not on the mobile terminal 11 but on a server on a network such as the Internet.

  In this case, the mobile terminal 11 accesses a server on the network and refers to the ambiguous word DB 22, task DB 23, purpose DB 25, importance DB 26, and position information DB 27.

  In the server, the mobile terminal 11 can refer to the latest data by appropriately updating the ambiguous word DB 22, task DB 23, purpose DB 25, importance DB 26, and position information DB 27.

  [Other examples of conversion of ambiguous words to specific numerical values]

  FIG. 17 is a diagram illustrating an example of an email transmitted (created) by the user A of the mobile terminal 11.

  The email M # 11 is an email sent from the user A (mobile terminal 11) to Mr. X in the same department as the user A of the company of the user A on March 9, 2011. It is a release issue, but it is expected to take a while. "

  E-mail M # 12 is an e-mail sent from user A to Mr. X on March 20, 2011, with the message “I will report on the matter that I am currently working on.” Yes.

  The mobile terminal 11 performs natural language analysis of the emails M # 11 and M # 12, and as a result, the work (task) of “internal release case” in the email M # 11 is completed by the email M # 12. It is recognized that Further, in the mobile terminal 11, the specific numerical value for the ambiguous word “for a while” in the mail M # 11 is changed from March 9, 2011 on the date of the mail M # 11 to March 2011 on the date of the mail M # 12. It is recognized that it is the 11th, which is the period up to the 20th of the month.

  In the portable terminal 11, the destinations (partner classes) of the emails M # 11 and M # 12 are people in the same section (section), and the topic of the message (the purpose of the task) is a topic for the inside of the company. Therefore, for example, a relatively low value is required as the importance of the task.

  Then, in the mobile terminal 11, the eleventh day, which is a specific numerical value for the ambiguous word “for a while”, the importance of the task, and the ambiguous word “for a while” in the email M # 11, is used as learning data. Learning is done.

  On the other hand, E-mail M # 21 is an e-mail sent from user A to customer Y on March 8, 2010. Is it? ”Is described.

  E-mail M # 22 is an e-mail sent from user A to Mr. Y on March 14, 2010, and contains the message "I'm sorry for being late than expected. I will deliver today." .

  The mobile terminal 11 performs natural language analysis of the emails M # 21 and M # 22. As a result, the “March 9 scheduled delivery” work (task) in the email M # 21 by the email M # 22. Is recognized as complete. Further, in the mobile terminal 11, the specific numerical value for the ambiguous word “for a while” in the mail M # 21 is from March 8, 2010 on the date of the mail M # 21 to March 2010 on the date of the mail M # 22. It is recognized that it is 6 days, which is the period until the 14th of the month.

  Further, in the mobile terminal 11, the destinations (partner class) of the mails M # 21 and M # 22 are human beings of the business partner (external), and the topic of the message (the purpose of the task) is a topic for the outside of the company. Therefore, for example, a relatively high value is required as the importance of the task.

  Then, the mobile terminal 11 uses, as learning data, 6 days that are specific numerical values for the ambiguous word “for a while”, the importance of the task, and the ambiguous word “for a while” in the email M # 21 as learning data. Learning is done.

  According to the discriminator in which learning is performed as described above, for the same ambiguous word “for a while”, depending on whether the mail destination (partner class) is in the section or outside the company, and the topic of the mail message Different specific numerical values are output depending on whether the (purpose of the task) is internal or external to the company.

  In the mobile terminal 11, the learning of the classifier is performed for each of the addresses M of the mail M # 11 and M # 12 and the mail M # 21 and M # 22. Can be prepared and done separately.

  FIG. 18 is a diagram illustrating an example of a blog comment (blog comment) created by the user A of the mobile terminal 11 and an email transmitted by the user A.

  The blog comment C # 1 is a comment describing that the user A visits the store AA, which is a tavern, at a drinking party with a colleague. The comment “Rating: 4.0, Purpose: Drinking party with a colleague, Review text: "Store AA is really delicious."

  Blog comment C # 2 is a comment describing that user A visited the restaurant BB, a French restaurant, on a date. The comment “Evaluation: 4.1, Purpose: Dating, Review: Store BB is OK. Is described.

  The mobile terminal 11 performs natural language analysis of the blog comments C # 1 and C # 2.

  Then, in the mobile terminal 11, for the blog comment C # 1, the specific numerical value for “very” in the review sentence is recognized as an evaluation of 4.0.

  Furthermore, in the mobile terminal 11, the blog comment C # 1 is a comment that targets a pub (partner class), and the pub is a store visited for the purpose of a drinking party with a colleague (the purpose of the task). Therefore, for example, a medium value is obtained as the importance of the task.

  Then, in the mobile terminal 11, the ambiguous word “very” in the blog comment C # 1, the importance of the task, and the evaluation value 4.0 that is a specific numerical value for the ambiguous word “very” are used as learning data to learn the classifier. Is done.

  Further, in the mobile terminal 11, for the blog comment C # 2, it is recognized that the specific numerical value for “maamaa” in the review sentence is an evaluation of 4.1.

  Further, in the mobile terminal 11, the blog comment C # 2 is a comment for a French restaurant (partner class), and the French restaurant is a store visited for the purpose of dating (task purpose). Therefore, for example, a relatively high value is obtained as the importance of the task.

  Then, the mobile terminal 11 uses, as learning data, the evaluation value 4.1, which is a specific numerical value for the ambiguous word “maama”, task importance, and the ambiguous word “maamaa” in the blog comment C # 2. Learning is done.

  The email M # 31 is an email sent from the user B (the mobile terminal 12) to the user A (the mobile terminal 11), and the message “This time, would you like to have a drink with the section person?” Is described.

  The mail M # 32 is a mail that the user A (the mobile terminal 11) is going to send to the user B (the mobile terminal 12) as a reply to the mail M # 31, and the message “XX” was very delicious. Is described.

  In the mobile terminal 11, for example, a medium value is obtained as the importance of the task based on the mails M # 31 and M # 32.

  Further, in the mobile terminal 11, the ambiguous word “very” included in the mail M # 32 is compared with the discriminator learned using the learning data obtained from the blog comments C # 1 and C # 2. The importance of the task is given as an input, and the discriminator outputs, for example, an evaluation of 4.0 is obtained as a specific numerical value for the ambiguous word “very”.

  Then, in the mobile terminal 11, as shown in FIG. 18, an evaluation 4.0, which is a specific numerical value for the ambiguous word “very”, is added to the mail M # 32 and transmitted.

  Note that the mobile terminal 11 of the user A can learn an identifier that converts an ambiguous word used by the user B into a specific numerical value by using an email exchanged with the user B (scalability). . In this case, miscommunication due to the user A misunderstanding the degree of the ambiguous word used by the user B can be prevented.

  In addition, the function of converting an ambiguous word into a specific numerical value in the mobile terminal 11 can be used for, for example, a compatibility diagnosis application that uses the closeness of a concept to a word as a key.

  That is, for example, in the mobile terminal 11, learning of the classifier FA that converts the ambiguous word used by the user A into a specific numerical value, and the classifier FB that converts the ambiguous word used by the user B into a specific numerical value. Keep learning and going.

  Thereafter, in the mobile terminal 11, the ambiguous word “little” included in “wait a moment”, which is an expression including the ambiguous word, is converted into a specific numerical value by each of the classifiers FA and FB.

  Then, the specific numerical value for the ambiguous word “a little” output from the discriminator FA and the specific numerical value for the ambiguous word “a little” output from the discriminator FB are, for example, 20 seconds and 30 minutes, respectively. On the other hand, if the mobile terminal 11 is far away, it is determined that the compatibility between the users A and B is poor in the mobile terminal 11.

  On the other hand, if the specific numerical value for the ambiguous word “a little” output from the classifier FA is close to the specific numerical value for the ambiguous word “a little” output from the discriminator FB, the mobile terminal 11 allows the users A and B to It is judged that compatibility with is good.

  [Description of computer to which this technology is applied]

  Next, the series of processes described above can be performed by hardware or software. When a series of processing is performed by software, a program constituting the software is installed in a general-purpose computer or the like.

  Thus, FIG. 19 shows a configuration example of an embodiment of a computer in which a program for executing the series of processes described above is installed.

  The program can be recorded in advance on a hard disk 105 or a ROM 103 as a recording medium built in the computer.

  Alternatively, the program can be stored (recorded) in the removable recording medium 111. Such a removable recording medium 111 can be provided as so-called package software. Here, examples of the removable recording medium 111 include a flexible disk, a CD-ROM (Compact Disc Read Only Memory), an MO (Magneto Optical) disk, a DVD (Digital Versatile Disc), a magnetic disk, and a semiconductor memory.

  In addition to installing the program from the removable recording medium 111 as described above, the program can be downloaded to the computer via a communication network or a broadcast network, and can be installed in the built-in hard disk 105. That is, for example, the program is wirelessly transferred from a download site to a computer via a digital satellite broadcasting artificial satellite, or wired to a computer via a network such as a LAN (Local Area Network) or the Internet. be able to.

  The computer incorporates a CPU (Central Processing Unit) 102, and an input / output interface 110 is connected to the CPU 102 via a bus 101.

  The CPU 102 executes a program stored in a ROM (Read Only Memory) 103 according to a command input by the user by operating the input unit 107 or the like via the input / output interface 110. . Alternatively, the CPU 102 loads a program stored in the hard disk 105 to a RAM (Random Access Memory) 104 and executes it.

  Thus, the CPU 102 performs processing according to the above-described flowchart or processing performed by the configuration of the above-described block diagram. Then, the CPU 102 outputs the processing result as necessary, for example, via the input / output interface 110, from the output unit 106, transmitted from the communication unit 108, and further recorded in the hard disk 105.

  The input unit 107 includes a keyboard, a mouse, a microphone, and the like. The output unit 106 includes an LCD (Liquid Crystal Display), a speaker, and the like.

  Here, in the present specification, the processing performed by the computer according to the program does not necessarily have to be performed in time series in the order described as the flowchart. That is, the processing performed by the computer according to the program includes processing executed in parallel or individually (for example, parallel processing or object processing).

  Further, the program may be processed by one computer (processor) or may be distributedly processed by a plurality of computers. Furthermore, the program may be transferred to a remote computer and executed.

  The embodiments of the present technology are not limited to the above-described embodiments, and various modifications can be made without departing from the gist of the present technology.

  Moreover, this technique can take the following structures.

[1]
An ambiguous word detection unit for detecting an ambiguous word having an ambiguous degree from input information inputted by a user;
From the input information, a task determination unit that determines a task related to the destination related to the ambiguous word;
An importance determination unit for determining the importance of the task;
Using the task importance and the ambiguous word as input, learning of a discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word, the importance, the ambiguous word, and the specific numerical value An information processing apparatus comprising: a conversion unit that converts the ambiguous word into the specific numerical value by giving the ambiguous word together with the importance as an input to the classifier obtained by using the identifier.
[2]
Generating a specific numerical value corresponding to the degree represented by the ambiguous word in response to completion of the task;
The information processing apparatus according to [1], further comprising: a learning unit that learns the classifier using the importance, the ambiguous word, and the specific numerical value.
[3]
An opponent class determination unit for determining an opponent class which is an opponent class of the task;
A purpose estimation unit for estimating the purpose of the task,
The information processing apparatus according to [1] or [2], wherein the importance level determination unit determines the importance level of the task based on the opponent class and the purpose of the task.
[4]
When the task is a waiting task that waits at a predetermined waiting time,
The information processing apparatus according to any one of [1] to [3], wherein the conversion unit converts the ambiguous word into a specific numerical value representing a degree of a delay time delayed by the waiting time, which is represented by the ambiguous word.
[5]
When the task is a waiting task that waits at a predetermined waiting time,
The conversion unit converts the ambiguous word into a specific numerical value of an offset with respect to a predicted value of the delay time, which is a degree of the delay time delayed by the waiting time, represented by the ambiguous word. The information processing apparatus according to any one of the above.
[6]
The information processing apparatus according to any one of [1] to [5], wherein the classifier is an SVM (Support Vector Machine), a neural network, or a linear regression model.
[7]
An opponent class determination unit that determines an opponent class that is an opponent class of the task;
The information processing apparatus according to [1] or [2], wherein the importance determination unit determines the importance of the task based on the opponent class.
[8]
A goal estimation unit for estimating the goal of the task;
The information processing apparatus according to [1] or [2], wherein the importance level determination unit determines the importance level of the task based on a purpose of the task.
[9]
From the input information input by the user, detect ambiguous words with an ambiguous degree,
From the input information, determine a task related to the destination related to the ambiguous word,
Determine the importance of the task,
Using the task importance and the ambiguous word as input, learning of a discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word, the importance, the ambiguous word, and the specific numerical value An information processing method including a step of converting the ambiguous word into the specific numerical value by giving the ambiguous word as an input together with the importance to the discriminator obtained by using the identifier.
[10]
An ambiguous word detection unit for detecting an ambiguous word having an ambiguous degree from input information inputted by a user;
From the input information, a task determination unit that determines a task related to the destination related to the ambiguous word;
An importance determination unit for determining the importance of the task;
Using the task importance and the ambiguous word as input, learning of a discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word, the importance, the ambiguous word, and the specific numerical value A computer functioning as a conversion unit for converting the ambiguous word into the specific numerical value by giving the ambiguous word together with the importance as an input to the discriminator obtained by using Program to let you.

  11, 12 Mobile terminal, 21 Dictionary storage unit, 22 Ambiguous word DB, 23 Task DB, 24 Partner class DB, 25 Objective DB, 26 Importance DB, 27 Location information DB, 31 Control unit, 32 User I / F, 33 Mail storage unit, 34 natural language analysis unit, 35 ambiguous word detection unit, 36 task determination unit, 37 partner class determination unit, 38 purpose estimation unit, 39 importance determination unit, 40 numerical value conversion unit, 41 classifier storage unit, 42 Transmission / reception unit, 43 position detection unit, 44 task achievement determination unit, 45 learning data generation unit, 46 learning data storage unit, 47 learning unit, 61 acceleration sensor, 62 behavior analysis unit, 63 behavior log storage unit, 64 numerical value correction unit, 101 bus, 102 CPU, 103 ROM, 104 RAM, 105 hard disk, 106 output unit, 107 Input unit, 108 communication unit, 109 drive, 110 input / output interface, 111 removable recording medium

Claims (10)

An ambiguous word detection unit for detecting an ambiguous word having an ambiguous degree from input information inputted by a user;
From the input information, a task determination unit that determines a task related to the destination related to the ambiguous word;
An importance determination unit for determining the importance of the task;
Using the task importance and the ambiguous word as input, learning of a discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word, the importance, the ambiguous word, and the specific numerical value An information processing apparatus comprising: a conversion unit that converts the ambiguous word into the specific numerical value by giving the ambiguous word together with the importance as an input to the classifier obtained by using the identifier. Generating a specific numerical value corresponding to the degree represented by the ambiguous word in response to completion of the task;
The information processing apparatus according to claim 1, further comprising: a learning unit that learns the classifier using the importance, the ambiguous word, and the specific numerical value. An opponent class determination unit for determining an opponent class which is an opponent class of the task;
A purpose estimation unit for estimating the purpose of the task,
The information processing apparatus according to claim 2, wherein the importance level determination unit determines the importance level of the task based on the opponent class and the purpose of the task. When the task is a waiting task that waits at a predetermined waiting time,
The information processing apparatus according to claim 3, wherein the conversion unit converts the ambiguous word into a specific numerical value representing a degree of a delay time that is delayed by the waiting time, which is represented by the ambiguous word. When the task is a waiting task that waits at a predetermined waiting time,
The information processing apparatus according to claim 3, wherein the conversion unit converts the ambiguous word into a specific numerical value of a correction value that corrects a predicted value of a degree of delay time delayed by the waiting time, which is represented by the ambiguous word. The information processing apparatus according to claim 3, wherein the classifier is an SVM (Support Vector Machine), a neural network, or a linear regression model. An opponent class determination unit that determines an opponent class that is an opponent class of the task;
The information processing apparatus according to claim 2, wherein the importance level determination unit determines the importance level of the task based on the opponent class. A goal estimation unit for estimating the goal of the task;
The information processing apparatus according to claim 2, wherein the importance level determination unit determines the importance level of the task based on a purpose of the task. From the input information input by the user, detect ambiguous words with an ambiguous degree,
From the input information, determine a task related to the destination related to the ambiguous word,
Determine the importance of the task,
Using the task importance and the ambiguous word as input, learning of a discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word, the importance, the ambiguous word, and the specific numerical value An information processing method including a step of converting the ambiguous word into the specific numerical value by giving the ambiguous word as an input together with the importance to the discriminator obtained by using the identifier. An ambiguous word detection unit for detecting an ambiguous word having an ambiguous degree from input information inputted by a user;
From the input information, a task determination unit that determines a task related to the destination related to the ambiguous word;
An importance determination unit for determining the importance of the task;
Using the task importance and the ambiguous word as input, learning of a discriminator that outputs a specific numerical value corresponding to the degree represented by the ambiguous word, the importance, the ambiguous word, and the specific numerical value A computer functioning as a conversion unit for converting the ambiguous word into the specific numerical value by giving the ambiguous word together with the importance as an input to the discriminator obtained by using Program to let you.

Images