JP4677051B1 - Conversation system and conversation sentence processing method - Google Patents

Abstract

The present invention realizes a conversation with a user without preparing in advance a DB for storing answers to conversation input sentences.
In a conversation system having a plurality of user terminal devices and a management server connected to the terminal device via a network, the management server responds to a sentence created by the user transmitted from the terminal device. Generate a unique sentence ID, analyze the sentence, determine the sentence type, check if there is a keyword that is characteristic in the sentence, and if there is a keyword, extract the keyword, and select the sentence as the answer A sentence processing unit to be created, and a sentence DB that stores acquired sentences and keywords in correspondence with the sentence IDs. If there is a keyword in the acquired sentence, the sentence processing unit searches the sentence DB, extracts sentences of the same sentence type including the same keyword, creates an answer using the extracted sentence, and the terminal device Send to.
[Selection] Figure 1

Description

  The present invention relates to a conversation system and a conversation sentence processing method, and more particularly, to a conversation system and a conversation sentence processing method in which a user connects a portable terminal to a management server and has a conversation by voice or sentences.

Various conversation systems have been proposed in which a user interacts with a character displayed on a display screen of a mobile terminal by voice or text while operating a mobile terminal such as a mobile phone.
For example, in Patent Document 1, in a pseudo-conversation system in which a user and a character freely have a pseudo-conversation, a content server having an answer database (DB) that stores answers corresponding to keywords is input from a client terminal. A system is disclosed in which a predetermined keyword is extracted from a sentence, a response database is searched, an answer corresponding to the extracted predetermined keyword is obtained, and distributed to a client terminal.

JP 2007-323551 A

  However, the technique described in Patent Document 1 must prepare an answer DB for storing answers in advance. Moreover, there are various keywords used in the conversation text, but it is necessary to prepare answer sentences corresponding to all the keywords in the answer DB. It takes a lot of work time to prepare all the answers. In some cases, the assumed keyword does not exist in the input sentence of the conversation. Even if a keyword exists in the input sentence, the answer sentence prepared in advance in the answer DB corresponds to the keyword, but the answer sentence does not always match the word (conversation sentence) used by the user. There is also a possibility.

  The present invention accumulates sentences input as conversations by the user in the DB and creates an answer using the sentences in the DB, so that it is not necessary to prepare in advance a DB for storing answers to the input sentences. Another object of the present invention is to provide a conversation system and conversation sentence processing method capable of realizing a conversation with a user.

The conversation system according to the present invention is preferably a conversation system having a management server connected via a network to a plurality of terminal devices of a user having an input device for inputting a sentence for conversation and a display for displaying the sentence. Because
The management server generates a unique sentence ID for a sentence created by a user transmitted from the terminal device, analyzes the sentence, determines a sentence type, and a keyword that is characteristic in the sentence If the keyword exists, the keyword is extracted, and a sentence processing unit that creates a reply sentence; the acquired sentence corresponding to the sentence ID; and the keyword A sentence DB (first sentence DB) to be stored;
In addition, when the keyword is included in the acquired sentence, the sentence processing unit searches the sentence DB, extracts sentences including the same keyword, and creates an answer using the extracted sentence. And a conversation system characterized by being transmitted to the terminal device.

In a preferred example, the management server further includes a second sentence DB that stores in advance a sentence that is an answer to the acquired sentence,
When there is no keyword in the acquired sentence, the sentence processing unit searches the second sentence DB, extracts a sentence as an answer, and transmits it to the terminal device.

Preferably, the management server further includes a user information DB for registering an ID for identifying a person having a personal relationship with the user corresponding to the user ID uniquely given to the user,
And said 1st sentence DB registers the user ID of the user who input the sentence corresponding to this sentence ID,
The sentence processing unit refers to the user information DB at the time of creating an answer to the acquired sentence, obtains a user ID of a user registered as a person having a personal relationship with the user, and The first sentence DB is searched for a human user ID, a sentence including the same keyword is extracted, an answer is created using the extracted sentence, and the answer is transmitted to the terminal device.

The conversational sentence processing method according to the present invention preferably has a management server connected via a network to a plurality of terminal devices having an input unit for inputting a sentence for conversation and a display unit for displaying the sentence. A conversation sentence processing method in a conversation system,
In the management server, generating a text ID unique to the text transmitted from the terminal device;
In the management server, the sentence transmitted from the terminal device is analyzed to determine the sentence type, and when there is a characteristic keyword in the sentence, the keyword is extracted;
In the management server, corresponding to the sentence ID, storing the acquired sentence and the keyword in a sentence DB (first sentence DB);
In the management server, as a result of analyzing the sentence, when there is a keyword in the sentence, the first sentence DB is searched, a sentence including the same keyword is extracted, and an answer is created using the extracted sentence Then, it is configured as a conversational sentence processing method characterized by having a step of transmitting to the terminal device.

In a preferred example, the management server stores in advance in the second sentence DB a sentence that is an answer to the acquired sentence,
When there is no keyword in the acquired sentence, the management server searches the second sentence DB, extracts a sentence as a reply, and transmits it to the terminal device.

Preferably, the management server further registers, in the user information DB, an ID for identifying a person having a personal relationship with the user, corresponding to the user ID uniquely given to the user,
And corresponding to the sentence ID, the user ID of the user who entered the sentence is registered in the first sentence DB,
The management server refers to the user information DB at the time of creating an answer to the input text, obtains the user ID of the user registered as a person of personal relation about the user, and obtains the user of the person of personal relation The first sentence DB is searched for the ID, a sentence including the same keyword is extracted, an answer is created using the extracted sentence, and the answer is transmitted to the terminal device.

  According to the present invention, it is possible to accumulate sentences input from a user's terminal device in a sentence DB and prepare an answer using the sentences in the sentence DB without preparing a DB for reply to conversation input sentences in advance. It becomes possible to create. This can save enormous work time for preparing answers in advance.

The figure which shows the whole structure of the conversation system by one Example. The figure which shows the structural example of the portable terminal in the conversation system by one Example. The figure which shows the example of sentence DB41 in the conversation system by one Example. The figure which shows the example of user information DB3 in the conversation system by one Example. The figure which shows the flowchart of the text process in the conversation system by one Example. The figure which shows the flowchart of the text process in the conversation system by another Example.

FIG. 1 shows an example of the overall configuration of a conversation system.
The conversation system includes a large number of mobile terminals 2 used by a user and a management server 1 connected to the mobile terminals 2 via a communication network 9.
The management server 1 is a computer that provides a conversation service to the user of the portable terminal 2. In this embodiment, the management server 1 displays animals and animated characters on the display of the portable terminal 3 of the user, It is realized that the user proceeds with the conversation by inputting the conversation text from the mobile terminal.

Although the hardware configuration of the management server 1 is not illustrated, the management server 1 has a processor (CPU), a memory, an external storage device such as a hard disk, an input device, a display device, etc., and executes a characteristic application program on the CPU. And realize the function to process sentences for conversation.
The mobile terminal 2 is a terminal such as a mobile phone, but may be a terminal device such as a PC (personal computer) normally used by a user. The hardware configuration of the mobile terminal 2 will be described later with reference to FIG.

In order to realize the conversation service, the management server 1 analyzes the conversation text acquired from the user information registration unit 102 for registering the information of the user of the portable terminal 2 and the portable terminal 2 and transmits it as an answer to the portable terminal. A sentence processing unit 103 that performs a process of creating a sentence to be performed. Moreover, it has user information DB3 (refer FIG. 4 ) which registers user information as a database (DB), and text DB4 (text DB41 and text DB42) which registers the text for conversation. The sentence DB 41 stores conversation sentences (input sentences) acquired from the portable terminal 2 and data of the analysis results (see FIG. 3 ). The sentence DB 41 can be said to be a user original sentence DB. The sentence DB 42 is prepared in advance by the service provider on the system, and stores simple sentences for answers in advance.

What is characteristic of the present invention is that if a characteristic word (keyword) is found as a result of analyzing the input sentence, the sentence DB 41 is searched for the keyword to obtain a sentence, and the sentence is used as an answer. Create a statement . On the other hand, if a keyword cannot be found in the input sentence, the sentence is extracted in advance from the sentence DB 42 and used as an answer sentence.
Here, the sentence processor 103 and text DB4 1 is the name of convenience of description, for example, may be referred to as a conversation part Toka conversation DB, the user and character performs a conversation using essentially the terminal device Other names may be used as long as they are intended.

FIG. 2 shows the configuration of the mobile terminal 2.
The portable terminal 2 includes a processor (CPU) 202 that executes various application programs, a memory 203 that stores data and programs, an input device 204, a display 205, and wireless communication that transmits and receives data to and from the communication network 9. The device 206 is configured. Unique identification information (user ID) is given to the user of the portable terminal 2, and the user ID is stored in the memory 203.
The user operates the input device 204 to input sentences necessary for conversation. The input text is processed by the CPU 202 and displayed on the display 205, and is sequentially transmitted to the management server 1. Note that the text for conversation is not limited to being input from the input device 204, but may be input from a voice input device. In this case, the CPU 202 of the mobile terminal converts the input speech into sentences sequentially while executing a speech recognition program for recognizing speech input from a speech input device (not shown), and displays it on the display device. indicate.

Next, the contents of each DB will be described with reference to FIGS.
FIG. 3 shows a configuration example of the sentence DB 41.
Sentence DB41 memorize | stores the data and the data by the analysis result which were transmitted and acquired from the portable terminal 2 under the process of the sentence process part 103. FIG. That is, a unique sentence ID assigned to each inputted sentence, an input sentence, a user ID unique to the user, a sentence sentence type to be stored, a keyword that is a characteristic word extracted from the sentence, and a kind of sentence If is a greeting, the type of greeting (morning, noon, night) is registered. The sentence type indicates one of the sentence types analyzed by the sentence processing unit 103 from among 17 predetermined (corpus classification) sentences. This will be described later.
Although the contents of the sentence DB 42 are not shown, the contents will be understood from the explanation of the generation of an answer sentence for the syntax analysis result of the input sentence described later. That is, as a result of corpus classification to be described later, the exemplified answer sentence when the feature word cannot be found in the input sentence is stored in the sentence DB 42.

FIG. 4 shows a configuration example of the user information DB 3.
The user information DB 3 registers personal information (for example, gender, date of birth, occupation, and address (not shown)) of a user who receives a service using the mobile terminal 2. As items characteristic of the present invention, user IDs uniquely assigned to each user, user names (may be nicknames), and friend data are registered. In the case of a plurality of friend items, the user sequentially adds and registers his / her friend. A friend is selected from those who can receive this conversation service, and a unique user ID is already assigned to each friend. In the friend column illustrated in FIG. 4, a surname is described, but a common name or a nickname may be used as long as the user ID can be specified. Alternatively, the user ID may be used instead of the last name on the DB.
As will be described later, the reason for registering the friend is to expand the search allowable range of the answer sentence so that the sentence processing unit 103 can select the answer including the input sentence from the registered friend. It is.

User information is registered in the user information DB 3 by first assigning a unique user ID to the user's personal information input and transmitted by the user's mobile terminal 2 under the processing of the user information registration unit 102. Is done. After that, when the user sequentially inputs and transmits necessary information, for example, the data related to the friend is updated to be added (or deleted).
In addition, since this user information DB registers each user's friend, you may call it friend DB. In some cases, a user password is given in addition to the user ID. In this case, the user password is also registered as one piece of user information.

Next, the conversation sentence processing operation in the management server 1 will be described with reference to the flowchart of FIG. This process is performed by the application program being executed by the CPU of the management server 1.
The user accesses a service providing website from the mobile terminal 2 and displays a character on the display 205 of the mobile terminal 2 to have a conversation. In order to receive a conversation service from this Web site, the user registers predetermined personal information in the user information DB 3 in advance. In addition, when accessing the Web site, the user ID and the user password must match.

  Now, after the mobile terminal 2 is connected to the Web site, the user inputs a text for conversation to the mobile terminal 2 and transmits it to the management server 1. The text transmitted from the portable terminal 2 includes a user ID. When the text processing unit 103 of the management server 1 acquires the input text transmitted from the mobile terminal 2, the text processing unit 103 generates a unique text ID and assigns it to the text (S501). The acquired text is temporarily stored in a memory (not shown).

Next, the sentence processing unit 103 performs syntax analysis of the input sentence (S502). The parsing of the input sentence will be described in detail later. To put it simply, the input sentence is first decomposed into parts of speech and subjected to primary classification, then the sentence is analyzed and classified into one of 17 predetermined types (secondary classification (also called corpus classification)). Determine the sentence type. Further, in this secondary classification, it is checked whether or not there is a word (keyword) that is a characteristic in the sentence. As a result, if there is a characteristic word, one or a plurality of characteristic words are extracted (S503).
Next, corresponding to the sentence ID generated previously, the acquired input sentence, user ID, sentence type and keyword as an analysis result are stored in the sentence DB 41 (S504).

  Next, the process proceeds to a process of creating a reply sentence (answer sentence) for the input sentence (S505). In this case, the sentence processing unit 103 confirms whether there is a feature word in the input sentence as a result of the previous syntax analysis (S506), and if there is a feature word as a result, searches the sentence DB 41 for the feature word. (S507). In the sentence DB 41, sentences and keywords when other users and the own user have conversed with the character in the past are stored together with the sentence ID. Referring to FIG. 3, for example, it is assumed that the personal name “Yumi” is extracted as a feature word from the input text of the user with the user ID “U012585” and stored in the keyword 1. If “Yumi” and “Yokohama” are extracted as feature words from the sentence “Yumi lives in Yokohama” with the sentence ID “P0005” this time, As a result of searching the sentence DB 41, a past sentence of the same user “I like Yumi” may be selected as an answer.

In addition, although not shown in FIG. 3, when a sentence “Yumi, I'm home today” is stored in the sentence DB 41 as a sentence from another user, the sentence “Yumi, There is a case where “I'm back home today” is selected (S508). The selected answer sentence is transmitted from the management server 1 to the user's portable terminal 2 and displayed on the display 205 (S509). The answer sentence is output as a character display or a voice of the character in accordance with the movement of the character being displayed on the screen, like a reply from the character. The user looks at the displayed answer sentence (or listens to the voice), and further continues the conversation and inputs the sentence.
As a result of searching the sentence DB 41, when a plurality of answer sentences hit the same keyword, for example, the current time is used as a seed of random numbers, and one of the corresponding answer sentences is selected according to the random number. An answer sentence shall be selected.

  On the other hand, in confirming whether there is a feature word in the input sentence (S506), if the feature word cannot be found, the sentence DB 42 is searched (S517), and the answer prepared in advance for each analysis syntax. A sentence is selected (S518). The selected answer sentence is transmitted to the user's portable terminal 2 in the same manner. Here, the rule for selecting an answer sentence from the sentence DB 42 is ambiguous. For example, when asked “Do you like **?”, You may decide specifically “I like it” or “I do n’t like it”. As another example, if it is an adjective sentence such as “** is cute”, an answer sentence that can be naturally connected to the adjective sentence is selected, and “cute” is applied to the template as the answer sentence. You may do it. In any case, the sentence DB 42 does not need to prepare a strict answer sentence that accurately matches the input sentence.

Next, with reference to the flowchart of FIG. 6, a sentence processing operation for conversation according to another example will be described. Compared with the above-described embodiment with reference to FIG. 5, the example of FIG. 6 is different in the processing of steps S621 to S623 (that is, the process of creating an answer sentence). The process for the difference will be described below.
The sentence processing unit 103 searches the user information DB 3 at the time of creating an answer sentence for the acquired sentence (S621), and acquires the user ID of the friend registered for the user ID of the user who is currently talking (S622). For example, the user (Mr. Suzuki) with the user ID “000002” acquires the user ID “U ***” of “Mr. Kudo”. Then, the entry of the user ID “U ***” and the user ID “000002” in the sentence DB 41 is searched for the feature word extracted previously. As a result of the search, an answer sentence including the characteristic word is selected.

  In the example of FIG. 5, all entries in the sentence DB 41 are searched, and sentences containing the same characteristic word are selected as answer sentences. May be selected. However, according to the example of FIG. 6, the search for the input sentence is only for the entry of the user himself / herself and the friend. A word is likely to be selected as an answer sentence. As a result, the answer sentence has a sense of familiarity, and the user can expect the conversation with the character to proceed more happily.

Next, the syntax analysis of the input sentence will be described.
The input sentence is analyzed through three stages 1 to 3.
1. Primary classification First, the part of speech is decomposed. Among them, “the presence or absence of one's name”, “the presence or absence of a noun”, “the presence or absence of a verb”, “the presence or absence of an imperative verb”, “the presence or absence of an adjective”, “whether the corpus itself is a questionable form” Make a decision (primary classification).

2. Secondary classification (corpus classification)
Based on the results of the primary classification, the following 17 types are classified (corpus classification).
(1) Incomprehensible example: “Asurari praises”
(2) Incomprehensible and calling example: "(Program name)-!"
(3) Sample sentences that cannot be understood: “What is **?”
(4) Example of a question for yourself: “(Name of program) is **?”
(5) Noun example: “* is *”
(6) Examples of noun sentence questions: “* is *?”
(7) Examples of questions about noun sentences for oneself: “Is ***?”
(8) Example adjective sentence: “* is cute”
(9) Adjective sentence questioning example: “Is * cute?”
(10) Example of adjective addressed to me: “(Program name) is cute”
(11) Adjective sentence questioning example addressed to me: “Is (program name) cute?”
(12) Verb example: “* ate *”
(13) Verb sentence addressed to me: "Does the name of the program eat *?"
(14) Verb sentence question example: “* eats *?”
(15) Verb sentence question example for oneself: “(name of program) eats *?”
(16) Verb sentence command example: “Eat!”
(17) Greeting example: “Good morning”
3. Processing of answer sentence by corpus classification (1) When a characteristic word is found in an incomprehensible sentence, a sentence including the characteristic word is randomly set as an answer sentence. If no feature word is found, the answer sentence is selected at random from all the sentences stored in the sentence DB 42.

(2) When a feature word is found in an incomprehensible and calling sentence, a sentence including the feature word is randomly selected as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “Eh? What is it?”
(3) When a characteristic word is found in an unintelligible question sentence, a sentence including the characteristic word is randomly selected as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “I do n’t know what it is…”
(4) When a characteristic word is found in the question sentence sentence addressed to itself, a sentence including the characteristic word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “Yeah!”
(5) When a characteristic word is found in a noun sentence sentence, a sentence including the characteristic word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “That's right!”
(6) When a feature word is found in a noun sentence question system sentence, a sentence including the feature word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “Oh ... I'm not sure”
(7) If a characteristic word is found in the noun sentence questioning sentence addressed to itself, a sentence including the characteristic word is randomly selected as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “Oh ... I'm not sure”
This process is the same as the above noun sentence question form, but is classified for future expansion.

(8) When a characteristic word is found in an adjective sentence sentence, a sentence including the characteristic word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence.
Example: If the input text is “Rabbit is cute, is n’t it?”

(9) When a characteristic word is found in an adjective sentence questioning sentence, a sentence including the characteristic word is randomly selected as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “I do n’t know.”
(10) When a characteristic word is found in the adjective sentence addressed to itself, a sentence including the characteristic word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence.
Example: If the input sentence is "(Program name) is cute, isn't it?"

(11) When a feature word is found in an adjective sentence questioned sentence addressed to itself, a sentence including the feature word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence.
Example: If the input text is "(Program name) is fun?"
Example: “Yeah, it ’s fun!”

(12) Verb sentence When a characteristic word is found in a sentence, a sentence including the characteristic word is randomly selected as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence. Example: “That's right!”.

(13) When a feature word is found in the verb sentence sentence addressed to itself, a sentence including the feature word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence.
Example: If the input sentence is "(Program name) eats sweets?", The answer is "Of course I will eat!"

(14) When a characteristic word is found in a verb sentence interrogative sentence, a sentence including the characteristic word is randomly selected as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence.
Example: “Huh? I ’m not sure”
(15) When a feature word is found in a question sentence related to a verb sentence addressed to itself, a sentence including the feature word is randomly selected as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence.
Example: "That's right!"

(16) When a feature word is found in a verb sentence command-type sentence, a sentence including the feature word is randomly set as an answer sentence. If a feature word is not found, it is selected from the sentence DB 42 as an answer sentence.
Example: If the input text is “Eat sweets!”, The response will be “Eat (user name)!

(17) Greetings Greetings are further subdivided into “morning greetings”, “daytime greetings”, “night greetings”, “greetings when returning home”, “greetings when parting”, and “first greetings” Is done.
In the case of greeting, when the input sentence is stored in the sentence DB 41, it is classified and stored in the morning, noon, and night. Then, when creating an answer, select the answer that matches the time and situation. For example, a morning greeting if it is from 6:00 am to 10:00 am, a day greeting if it is from 10 am to 6:00 pm, and a night greeting from 6:00 pm to 6:00 am the next day. select. For special greetings such as “Nice to meet you”, an answer sentence such as “Nice to meet you” may be prepared in the sentence DB 41 and selected when creating the answer sentence.

  In the above description, there are various expressions such as “when a feature word is found in a sentence, a sentence including the feature word is randomly selected as an answer sentence”. In this method, for example, the current time is used as a seed of random numbers, and randomly selected from a plurality of corresponding answer sentences using random numbers.

As mentioned above, although the Example of this invention was described, this invention is not limited to the said Example, It can implement by various deformation | transformation and application.
For example, in the above embodiment, when no feature word is found in the input sentence, the sentence is selected from the sentence DB 42 as the answer sentence. However, the sentence DB 42 for storing the answer sentence in advance is not necessarily required. For example, when a feature word is not found in the input sentence, a reply sentence “I don't know” may be used regularly. In this case, a simple single sentence “I don't know” can be generated in the program without being stored in the sentence DB 42.

  As yet another example in the case where no feature word is found in the input sentence, the sentence DB 41 is searched using the user ID of the input sentence as a key, and the sentence input by the user in the past is selected as an answer sentence. Also good. In this case, for example, when the input sentence is “Eat sweets!”, The same sentence “Eat sweets!” May be selected as an answer sentence. As yet another example, an answer sentence of “Good morning” may be selected for an input sentence of “Good morning”.

  Further, as a modification of the text processing shown in FIG. 6, searching for a special entry in the text DB 41 is not limited to the user's friend. For example, it is possible to search for an entry of a user ID of a person who has a special personal relationship such as a user's parents, siblings, or children. In this case, in the user information DB 3 of FIG. 4, not only the user's friends but also user IDs of persons having a special personal relationship are registered.

1: Management server 2: Mobile terminal 9: Network
102: User information registration unit 103: Text processing unit 3: User information DB 4: Text DB 41: Text DB1 42: Text DB2
202: CPU 203: Memory 204: Input device 205: Display device 206: Wireless communication device

Claims (5)

A conversation system having a management server connected via a network to a plurality of terminal devices of a user having an input device for inputting a text for conversation and a display for displaying the text,
The management server
A unique sentence ID is generated for the sentence created by the user transmitted and acquired from the terminal device, and the sentence is further decomposed into parts of speech to perform primary classification, and then the primary classified sentence is determined in advance. The sentence type is determined by performing a secondary classification as to which of the multiple types of corpus classifications, and the sentence type is determined, and there are characteristic keywords in the secondary classification of the sentence. And if there is the keyword, the keyword is extracted, and further a sentence processing unit for creating a reply sentence;
And said sentences are parsed by the text processing unit, and said keywords extracted, and the document type determined by the parsing, a first sentence DB for storing in response to said text ID,
Corresponding to a plurality of types of the corpus, the second sentence DB for storing the answer sentence in advance ,
The sentence processing unit, the syntax analysis result, if the keyword is present in the secondary Classified said sentence extracting a sentence including the same said keyword by searching the first sentence DB, extracted the create a sentence for the answer by using the text into a text sent to the terminal device,
The sentence processing unit searches the second sentence DB and corresponds to the corpus classification determined by the secondary classification when the keyword is not found in the secondary classified sentence as a result of the syntax analysis. A conversation system, wherein the sentence for answering is extracted and transmitted to the terminal device . The management server further includes a user information DB for registering an ID for identifying a person having a personal relationship with the user corresponding to the user ID uniquely given to the user,
And said 1st sentence DB registers the user ID of the user who input the sentence corresponding to this sentence ID,
The sentence processing unit refers to the user information DB when the keyword is included in the sentence that is secondarily classified as a result of syntactic analysis of the sentence, and is registered as a person having a personal relationship with the user. The first sentence DB is searched for the user ID of the person who has the personal relationship, the sentence including the same keyword is extracted, and an answer is created using the extracted sentence. The conversation system according to claim 1, wherein the conversation system is transmitted to the terminal device. A conversation sentence processing method in a conversation system having a management server connected via a network with a plurality of terminal devices having an input device for inputting sentences for conversation and a display device for displaying sentences,
In the management server, a unique sentence ID is generated with respect to the sentence created by the user transmitted and acquired from the terminal device, and the sentence is further decomposed to perform primary classification, and then the primary classification is performed. The sentence type is determined by performing a secondary classification to determine which of a plurality of types of corpus classifications the sentence is in advance, and the sentence type is determined, and features are included in the secondary classified sentence Checking if there is a keyword to be extracted, and if there is the keyword, extracting the keyword;
In the management server, and the sentence is parsed by the text processing, and the keyword extracted, and the document type determined by the parsing, corresponds to the sentence ID and stored in the first sentence DB And steps to
Result of parsing of the sentence, if the keyword is present in the secondary sorted the sentence, by searching the first sentence DB, extracting a sentence including the same said keyword, the extracted the sentence Using to create an answer and send it to the terminal device;
If the keyword is not found in the secondary classified sentence as a result of syntactic analysis of the sentence, the second sentence DB in which answer sentences corresponding to a plurality of types of the corpus classification are stored in advance is searched. And extracting the response text corresponding to the corpus classification determined by the secondary classification and transmitting it to the terminal device . The management server further includes a user information DB for registering an ID for identifying a person having a personal relationship with the user corresponding to the user ID uniquely given to the user,
And said 1st sentence DB registers the user ID of the user who input the sentence corresponding to this sentence ID,
As a result of syntactic analysis of the sentence, when the keyword is included in the sentence that is secondarily classified, with reference to the user information DB, the user ID of the user registered as a person having a personal relationship with the user The first sentence DB is searched for the user ID of the person having the personal relationship, the sentence including the same keyword is extracted, and an answer is created using the extracted sentence and transmitted to the terminal device. The conversational sentence processing method according to claim 3, wherein: In the sentence processing of the management server, when a plurality of sentences including the same keyword are extracted as a result of searching the first sentence DB, one sentence is selected from among the plurality of sentences using a random number, 5. A method for processing a conversational sentence according to claim 3 or 4, wherein a sentence for answering is created.

Images