WO2019012872A1 - 自動応答サーバー装置、端末装置、応答システム、応答方法、およびプログラム - Google Patents
自動応答サーバー装置、端末装置、応答システム、応答方法、およびプログラム Download PDFInfo
- Publication number
- WO2019012872A1 WO2019012872A1 PCT/JP2018/021857 JP2018021857W WO2019012872A1 WO 2019012872 A1 WO2019012872 A1 WO 2019012872A1 JP 2018021857 W JP2018021857 W JP 2018021857W WO 2019012872 A1 WO2019012872 A1 WO 2019012872A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- response
- chat
- input
- knowledge data
- fragment
- Prior art date
Links
- 230000004044 response Effects 0.000 title claims abstract description 513
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000012545 processing Methods 0.000 claims abstract description 106
- 230000008569 process Effects 0.000 claims abstract description 41
- 230000005540 biological transmission Effects 0.000 claims abstract description 34
- 239000012634 fragment Substances 0.000 claims description 154
- 238000013500 data storage Methods 0.000 claims description 67
- 238000007726 management method Methods 0.000 claims description 9
- 230000003287 optical effect Effects 0.000 claims description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 14
- 230000006870 function Effects 0.000 description 20
- 238000010586 diagram Methods 0.000 description 15
- 238000012986 modification Methods 0.000 description 15
- 230000004048 modification Effects 0.000 description 15
- 238000004458 analytical method Methods 0.000 description 7
- 238000004891 communication Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 239000000203 mixture Substances 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 230000015654 memory Effects 0.000 description 3
- 238000003825 pressing Methods 0.000 description 3
- 238000006467 substitution reaction Methods 0.000 description 3
- 125000002066 L-histidyl group Chemical group [H]N1C([H])=NC(C([H])([H])[C@](C(=O)[*])([H])N([H])[H])=C1[H] 0.000 description 2
- 210000000988 bone and bone Anatomy 0.000 description 2
- 238000010411 cooking Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012015 optical character recognition Methods 0.000 description 2
- 235000015277 pork Nutrition 0.000 description 2
- 235000008694 Humulus lupulus Nutrition 0.000 description 1
- 206010020751 Hypersensitivity Diseases 0.000 description 1
- 208000026935 allergic disease Diseases 0.000 description 1
- 230000007815 allergy Effects 0.000 description 1
- 230000009118 appropriate response Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000007596 consolidation process Methods 0.000 description 1
- 239000002537 cosmetic Substances 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/42—Systems providing special services or facilities to subscribers
- H04M3/487—Arrangements for providing information services, e.g. recorded voice services or time announcements
- H04M3/493—Interactive information services, e.g. directory enquiries ; Arrangements therefor, e.g. interactive voice response [IVR] systems or voice portals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/02—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail using automatic reactions or user delegation, e.g. automatic replies or chatbot-generated messages
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
- G06N5/041—Abduction
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification techniques
- G10L17/22—Interactive procedures; Man-machine interfaces
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L51/00—User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail
- H04L51/04—Real-time or near real-time messaging, e.g. instant messaging [IM]
Definitions
- the present invention relates to an automatic response server device, a terminal device, a response system, a response method, and a program.
- Priority is claimed on Japanese Patent Application No. 2017-138423, filed July 14, 2017, the content of which is incorporated herein by reference.
- chat system As a representative.
- chat systems are often used for communication between individuals, but may be used for business purposes.
- the business operator can use the chat system for the purpose of publicity, advertisement, and the purpose of some customer service.
- some customer services are activities for responding to inquiries from customers and providing information to customers.
- chat service technology By combining the chat service technology and the artificial intelligence technology described above, it is possible to promptly respond to various questions the customer has and requests for providing information.
- Patent Document 1 describes a method of chat between human and machine which obtains a user's intention based on an input signal and outputs a signal corresponding to the user's intention by using artificial intelligence.
- the present invention has been made based on the above problem recognition, and can reduce the investment in a computer required for learning processing in artificial intelligence, an automatic response server device, a terminal device, and a response system. , A response method, and a program.
- a response system is a response system including an automatic response server device and a plurality of terminal devices, wherein the terminal device is configured to perform the automatic communication in a chat
- a chat pattern generation unit that generates information of a chat pattern that is a pattern of an input fragment that is input to a response server device and a response fragment that is a response from the automatic response server device to the input fragment;
- the response fragments included in the chat pattern generated by the chat pattern generation unit Response knowledge data transmission unit for transmitting response knowledge data to the automatic response server device, and input interpretation knowledge data for transmitting the input interpretation knowledge data output from the learning processing unit to the automatic response server device And a response knowledge data storage unit for storing the response knowledge data sent from the response knowledge data sending unit of the terminal device.
- An input interpretation knowledge data storage unit for storing the input interpretation knowledge data transmitted from the input interpretation knowledge data transmission unit, an input text in a chat, and the input stored in the input interpretation knowledge data storage unit Of the response knowledge data stored in the response knowledge data storage unit based on the interpretation knowledge data,
- An inference engine unit that estimates a response fragment corresponding to the input text and outputs a chat response by reading out response knowledge data corresponding to the estimated response fragment from the response knowledge data storage unit; It is characterized by having.
- One aspect of the present invention further comprises, in the response system described above, a prototype database server device provided with a prototype storage unit for storing a prototype that is a prototype of the chat pattern, wherein the chat pattern generation unit Information of the chat pattern is generated based on the prototype acquired from the prototype database server device.
- One embodiment of the present invention is the response system described above, wherein the prototype database server device comprises a field information management unit that manages information of the field to which the prototype belongs, and the prototype is associated with the field Be managed.
- the automatic response server device further includes an access information generation unit that generates image information including the access information that is optically readable.
- the access information acquisition unit acquires image information including the optically readable access information generated by the access information generation unit.
- the inference engine unit is connected via a chat server device for providing a chat service between users via the chat server device in general
- the chat with the user terminal device is executed, and the input text in the chat is sent from the general user terminal device to the inference engine unit via the chat server device.
- the response in the chat is output by the inference engine unit, and sent to the general user terminal device via the chat server device.
- the input text in a chat is transmitted to the inference engine unit via the chat server device, and the response output from the inference engine unit is
- the terminal may further comprise a general user terminal device that receives via the chat server device.
- One aspect of the present invention is that, in the response system described above, the access information acquired by the access information acquisition unit in the terminal device is read by an optical reader and based on the read access information The chat is accessed, and the input text in the chat is transmitted to the inference engine unit via the chat server device, and the response output from the inference engine unit is received via the chat server device And a general user terminal device.
- One aspect of the present invention is a response method using an automatic response server device and a plurality of terminal devices, wherein, in the terminal device, a chat pattern generation unit sends the automatic response server device in a chat. Information of a pattern of an input fragment which is an input and a response fragment which is a response from the automatic response server device to the input fragment is generated, and a learning processing unit generates the chat pattern based on the chat pattern generated by the chat pattern generation unit.
- the input interpreting knowledge data transmitting unit transmits to the bar apparatus, and the input interpreting knowledge data output from the learning processing unit is transmitted to the automatic response server apparatus.
- the knowledge data storage unit stores the response knowledge data transmitted from the response knowledge data transmission unit of the terminal device, and the input interpretation knowledge data storage unit transmits the input interpretation knowledge data of the terminal device.
- the inference engine unit responds based on the input text in the chat and the input interpretation knowledge data stored in the input interpretation knowledge data storage unit.
- the response fragment corresponding to the input text in the chat among the response knowledge data stored in the knowledge data storage unit Estimated, it outputs the response of the chat by reading the response for the knowledge data corresponding to the estimated the response fragment from the response for the knowledge data storage unit, a response method characterized by.
- One aspect of the present invention is information of a chat pattern which is a pattern of an input fragment which is an input to an automatic response server apparatus in chat and a response fragment which is a response from the automatic response server apparatus to the input fragment.
- a learning process of the relationship between the input text corresponding to the input fragment and the response fragment is performed based on the chat pattern generating unit to be generated and the chat pattern generated by the chat pattern generating unit, and the result of the learning process
- a learning processing unit for outputting input interpretation knowledge data obtained as a response, and a response for transmitting response knowledge data based on the response fragment included in the chat pattern generated by the chat pattern generation unit to the automatic response server device
- Knowledge data transmission unit, and the input interpretation knowledge data output from the learning processing unit A terminal device which is characterized by comprising an input interpretation for knowledge data transmitter for transmitting data to the automatic answering server device.
- One aspect of the present invention is response knowledge data that stores, as response knowledge data, a response fragment that is a response to an input fragment that is an input in a chat based on a chat pattern generated in an external terminal device Storage unit and knowledge data for input interpretation generated by learning processing in the terminal device, regarding the relationship between the input text corresponding to the input fragment in the chat pattern and the response fragment corresponding to the input fragment
- An input interpretation knowledge data storage unit for storing input interpretation knowledge data obtained as a result of learning processing, an input text in a chat, and the input interpretation knowledge data stored in the input interpretation knowledge data storage unit Of the response knowledge data stored in the response knowledge data storage unit.
- An inference engine unit that estimates a response fragment corresponding to an input text in a chat, and outputs a chat response by reading out response knowledge data corresponding to the estimated response fragment from the response knowledge data storage unit;
- An automatic response server apparatus comprising:
- One embodiment of the present invention is a chat pattern in which a computer is a pattern of an input fragment that is input to an automatic response server device in chat and a response fragment that is a response from the automatic response server device to the input fragment. And a learning process of the relationship between the input text corresponding to the input fragment and the response fragment, based on the chat pattern generating unit generating the information of the above, and the chat pattern generated by the chat pattern generating unit; A learning processing unit that outputs input interpretation knowledge data obtained as a result of the processing, and response knowledge data based on the response fragment included in the chat pattern generated by the chat pattern generation unit to the automatic response server device
- the response knowledge data transmission unit to be transmitted, and the previous output from the learning processing unit Is a program for causing an input interpretation for knowledge data as an input interpretation for knowledge data transmission unit to be transmitted to the automatic answering server device.
- One aspect of the present invention is response knowledge data for storing, as response knowledge data, a response fragment that is a response to an input fragment that is an input in a chat, based on a chat pattern generated in an external terminal device Storage unit and knowledge data for input interpretation generated by learning processing in the terminal device, regarding the relationship between the input text corresponding to the input fragment in the chat pattern and the response fragment corresponding to the input fragment
- a computer comprising an input interpretation knowledge data storage unit for storing input interpretation knowledge data obtained as a result of learning processing, an input text in a chat, and the input stored in the input interpretation knowledge data storage unit Stored in the response knowledge data storage unit based on the interpretation knowledge data
- the response fragment corresponding to the input text in the chat is estimated, and the response knowledge data corresponding to the estimated response fragment is read out from the response knowledge data storage unit.
- the present invention it is possible to learn artificial intelligence without putting a heavy load on learning processing on the automatic response server device, and to realize an automatic response of chat based on the learning result.
- the user of the member terminal can easily construct a response system customized for his / her business by creating a chat pattern.
- a user of a member terminal downloads a previously prepared prototype of a chat pattern from a prototype database server device. Then, the user of the member terminal device creates a chat pattern by setting a response (for example, a response to the question) to an input (for example, a question) based on the prototype.
- the chat pattern prototype includes, for example, examples of representative input and similar input.
- FIG. 1 is a block diagram showing a schematic functional configuration of a response system 1 according to a first embodiment of the present invention. It is a schematic diagram showing an example of composition of a chat fragment (input fragment) in the embodiment. It is a schematic diagram showing an example of composition of a chat fragment (input fragment) in the embodiment. It is a schematic diagram showing an example of composition of a chat fragment (response fragment) in the embodiment. It is a schematic diagram showing an example of composition of a chat fragment (response fragment) in the embodiment. It is the schematic which shows the 1st example of the chat pattern used in the embodiment. It is the schematic which shows the 2nd example of the chat pattern used in the embodiment.
- FIG. 14 is a schematic view showing an example of a screen for registering a chat pattern in the member terminal 3 according to the same embodiment.
- the member terminal device 3 by the same embodiment it is a schematic diagram showing an example of a screen for editing and registering a synonym dictionary.
- FIG. 14 is a schematic view showing an example of a screen for editing and registering a synonym dictionary.
- FIG. 17 is a schematic view showing an example of screen display of a chat performed between a general user terminal device 7 and a virtual user in the automatic response server device 2 according to the same embodiment, and the general user terminal device 7 prints It also shows the flow of accessing a specific chat service in the chat server device 5 by reading the stored access information.
- It is a block diagram showing a schematic functional configuration of a response system 101 according to a second embodiment of the present invention. It is a block diagram showing a schematic functional configuration of a response system 101 according to a third embodiment of the present invention.
- FIG. 1 is a block diagram showing a schematic functional configuration of a response system according to the first embodiment.
- the response system 1 includes an automatic response server device 2, a member terminal device 3 (terminal device), a prototype database server device 4, a chat server device 5, and a general user terminal device 7. It consists of The automatic response server device 2, the member terminal device 3, the prototype database server device 4, and the chat server device 5 can communicate with each other via a wired or wireless communication line or the like. In this communication, for example, the Internet Protocol (Internet Protocol) is used.
- Internet Protocol Internet Protocol
- each device shown in FIG. 1 is realized, for example, using an electronic circuit.
- Each function may be internally provided with storage means such as a semiconductor memory or a magnetic hard disk drive, as necessary.
- Each function may be realized by a computer and software.
- the automatic response server operator holds and operates the automatic response server device 2 and the prototype database server device 4.
- a member who uses the service of the automatic response server holds and uses the member terminal 3.
- a part of the software operating in the member terminal 3 may be provided by the automatic response server operator.
- the chat service operator holds and operates the chat server device 5.
- the chat service provider may be the same as the above-mentioned automatic response server provider.
- a general user accesses the chat server device 5 using the general user terminal device 7 and receives a service.
- the general user may be an unspecified number.
- the outline of each device in the response system 1 is as follows.
- the automatic response server device 2 is a device that has implemented the function of receiving data input via the chat service and returning an appropriate response to the input.
- the automatic response server device 2 is operated as a server for a plurality of members.
- the members are, for example, a store, a restaurant, a financial institution, and other businesses.
- the member terminal 3 is a terminal used by a member.
- the member terminal device 3 is also simply referred to as a "terminal device".
- the member terminal device 3 is realized by, for example, a PC (personal computer), a tablet terminal device, a smartphone or the like.
- the member terminal 3 obtains a chat pattern prototype from the prototype database server device 4.
- the member terminal 3 has a function to edit and register the chat pattern based on the acquired prototype.
- the member terminal device 3 performs a learning process based on the registered data of the chat pattern, and provides the automatic response server device 2 with the knowledge data which is the learning result.
- the knowledge data provided by the member terminal 3 to the automatic response server 2 includes input interpretation knowledge data and response knowledge data. These knowledge data will be described later.
- the prototype database server device 4 holds prototype data of the chat pattern.
- the prototype database server device 4 is realized using, for example, a server type computer and software of a database management system (DBMS).
- DBMS database management system
- the prototype database server device 4 also manages the field tree.
- the chat pattern and the field tree will be described in detail later with reference to the drawings.
- the chat server device 5 provides a chat service.
- the chat server device 5 is realized using, for example, a server type computer and software for implementing a chat service.
- the chat server device 5 provides a function for performing chat between two users.
- the chat server device 5 enables implementation of a chat between the general user terminal device 7 and the virtual user executed by the automatic response server device 2.
- the chat using the chat server device 5 may be performed using data other than text (multimodal data), but here, a chat performed using only text data will be described.
- the chat server device 5 realizes chat between the user A and the user B.
- the text data input by user A and the text data input by user B are displayed in a chat room (virtual room) shared by both users.
- the user A and the user B can display a text dialogue exchanged in this chat room on a screen or the like.
- one of the chat users is a virtual user on the automatic response server device 2.
- the program of the automatic response server device 2 acquires text data in a chat room or transmits text data to the chat room via an API (application program interface). Thereby, it is possible to chat between the general user terminal device 7 (user A) and a virtual user (user B, program) on the automatic response server device.
- the general user terminal device 7 is a terminal device used by a general user.
- the general user terminal device 7 is realized by, for example, a PC, a tablet terminal device, a smartphone, a wearable terminal, or the like.
- the general user terminal device 7 incorporates client software for using the chat service.
- This client software is, for example, a general-purpose web browser or dedicated chat client software (application).
- the automatic response server device 2 shown in FIG. 1 includes an inference engine unit 21, an input interpretation knowledge data storage unit 22, a response knowledge data storage unit 23, an input interpretation knowledge data acquisition unit 24, and response knowledge data. And an acquisition unit 25.
- These units are realized, for example, using a computer program and a memory. The functions of these units are as follows.
- the inference engine unit 21 has a function of executing chat with the general user terminal device 7 via the chat server device 5. In other words, the inference engine unit 21 has a function of operating as a virtual user who chats with the general user terminal device 7. Specifically, the inference engine unit 21 receives the input text T input from the chat server device 5. This input text T input is text transmitted from the general user terminal device 7 side. Next, the inference engine unit 21 interprets the input based on the received input text T input and estimates a response most suitable for the interpreted input. The inference engine unit 21 returns the estimated optimum response to the general user terminal. Specifically, when the input text T input is given, the inference engine unit 21 estimates the correct answer represented by the following equation (1).
- i is the index number of the response RESP.
- RESP i is a response having an index number i.
- ⁇ is a group of parameter values for calculating P (RESP i
- T input ; ⁇ ) is the conditional probability of RESP i given T input .
- T input ; ⁇ ) depends on the parameter value group ⁇ .
- the parameter value group ⁇ is appropriately adjusted by learning processing using teacher data. That is, the parameter value group ⁇ is knowledge data (knowledge data for input interpretation) in artificial intelligence technology (machine learning technology).
- the input interpretation knowledge data is stored in the input interpretation knowledge data storage unit 22.
- the inference engine unit 21 may be realized using artificial intelligence technology.
- the inference engine unit 21 is realized using a rule base and a multi-layered neural network.
- it may be realized using another form in the artificial intelligence field.
- the above-mentioned parameter value group ⁇ includes values of weights in the connection between neurons.
- the inference engine unit 21 When a correct answer that is optimal is estimated based on the input text T input , the inference engine unit 21 generates data for generating an output text according to the estimated optimal response from the response knowledge data storage unit 23. read out.
- the response knowledge data is, for example, the response text itself.
- the response knowledge data is a set of response knowledge data including a variable and a method for acquiring a specific value of the variable.
- the inference engine unit 21 Based on the read response knowledge data, the inference engine unit 21 generates text data of the response.
- the text data of the response may include variables for padding.
- the inference engine unit 21 can appropriately read out the specific value of the variable from a database (not shown) by executing the above method.
- the inference engine unit 21 calculates the specific value of the variable ⁇ x>, the balance of the account of the user (whose authentication process is performed separately with respect to being an account holder). read out. Procedures for acquiring data of specific values of variables can be described as methods. The method will be described later. Then, the inference engine unit 21 outputs, as a response, a text in which the variable ⁇ x> is replaced with a specific numerical value. As a specific example, the text "The balance of the account is 12,345,678 yen.” Is output. The inference engine unit 21 returns the generated response text to the general user terminal device 7 via the chat server device 5.
- the inference engine unit 21 uses the response feature stored in the response knowledge data storage unit 23 based on the features of the input in the chat and the input interpretation knowledge data stored in the input interpretation knowledge data storage unit 22. Among knowledge data, an optimal one (optimum response fragment) is estimated, and a chat output text is output based on the estimated response fragment.
- the inference engine unit 21 extracts features of input in chat from input text in chat.
- the feature of input in chat is a feature word group in input text.
- features of syntax in input text or other features may be used as features of input in chat.
- the feature of the syntax in the input text is information representing the relationship between each node in the syntax tree.
- the information is, for example, information of mutual positional relationship or mutual distance in the syntax tree between a specific word and another specific word. That is, the inference engine unit 21 stores the response knowledge stored in the response knowledge data storage unit 23 based on the input text in the chat and the input interpretation knowledge data stored in the input interpretation knowledge data storage unit 22. Among the data, an optimal response fragment corresponding to the input text in the chat is estimated, and a response of the chat is output by reading out the response knowledge data corresponding to the estimated response fragment from the response knowledge data storage unit 23 .
- the chat pattern representing the chain relationship between the input and the response may have various forms in addition to the simple “input ⁇ response” pair. An example of the form of the chat pattern will be described later.
- the inference engine unit 21 can execute chats with a plurality of general user terminal devices 7 in parallel (concurrently) independently of each other. In other words, the inference engine unit 21 can conduct conversations in many chat rooms concurrently.
- the inference engine unit 21 may internally have a database for accumulating and storing all the records of chats exchanged between the general user terminal device 7 and itself.
- this database when the chat is performed only with text, the text data sent from the general user terminal device 7 to the inference engine unit 21 and the general user terminal device 7 from the inference engine unit 21. Store all of the text data sent to the side.
- the chat includes multimodal data other than text data (for example, when audio data, still image data, moving image data, and other data are included), all of the multimodal data is in the inference engine unit 21.
- the database of the inference engine unit 21 may be provided in a device outside the automatic response server device 2. As described above, by keeping the chat record, the automatic response server device 2 can utilize the data stored in the database as big data as an analysis target.
- the input interpretation knowledge data storage unit 22 stores knowledge data to be used by the artificial intelligence included in the inference engine unit 21.
- An example of the knowledge data is the aforementioned parameter value group ⁇ .
- the knowledge data stored in the input interpretation knowledge data storage unit 22 corresponds to the input text passed from the general user terminal device 7 to estimate the optimum response fragment in the set of response fragments. It is a parameter value group.
- the input interpretation knowledge data storage unit 22 stores the input interpretation knowledge data transmitted from the input interpretation knowledge data transmission unit 35 of the member terminal device 3.
- the response knowledge data storage unit 23 stores data of response fragments.
- the data of the response fragment includes, for example, information for uniquely identifying the response fragment, response text (which may include a variable), and a method for resolving the variable if the response text includes a variable. including.
- the response knowledge data storage unit 23 stores the response knowledge data transmitted from the response knowledge data transmission unit 33 of the member terminal device 3. Specific examples of chat patterns that are sources of knowledge data (knowledge data for input interpretation and knowledge data for response) will be described later with reference to FIGS. 2 to 7.
- the input interpretation knowledge data acquisition unit 24 acquires input interpretation knowledge data from the member terminal device 3 and writes the input interpretation knowledge data in the input interpretation knowledge data storage unit 22.
- the response knowledge data acquisition unit 25 acquires response knowledge data from the member terminal device 3 and writes the response knowledge data into the response knowledge data storage unit 23. Details of processing on the member terminal 3 side and timing of updating (writing) knowledge data and chat patterns will be described later.
- the prototype database server device 4 shown in FIG. 1 includes a prototype storage unit 41 and a field tree management unit 42.
- the prototype storage unit 41 stores prototype data.
- a prototype is data that is a prototype of a chat pattern.
- the prototype is sample data having the same format as the chat pattern.
- the prototype is associated with the field identifier for identifying one field in the field tree managed by the field tree management unit 42.
- the field tree management unit 42 (field information management unit) stores and manages data of the field tree.
- the domain tree includes a label representing a domain and an identifier for uniquely identifying the domain.
- a discipline tree manages disciplines in a tree structure. An example of the domain tree will be described later with reference to the drawings. That is, the field tree management unit 42 stores and manages information of the field to which the prototype belongs.
- a chat pattern is a chain of chat fragments.
- chat fragments There are two types of chat fragments: input fragments and response fragments.
- the input fragment may be simply referred to as "input” and the response fragment may be referred to simply as "response”.
- the input fragment is an input to the automatic response server device 2.
- the response fragment is a response output from the automatic response server device 2. That is, the chat pattern is data representing a pattern composed of an input to the automatic response server device 2 and a response from the automatic response server device 2 to the input. That is, the chat pattern is data expressed as a chain of input fragments and response fragments.
- the chat pattern is data representing a chat pattern performed by the automatic response server device 2 with the general user terminal device 7.
- the chat pattern is a pattern that represents the relationship between the input to the automatic response server device 2 and the response from the automatic response server device 2.
- An input included in the chat pattern may be associated with a feature word group for representing a feature of the input.
- the chat pattern can be expressed as data of an appropriate format. As an example, the chat pattern can be expressed in XML (extensible markup language).
- FIG. 2A to 2D are schematic diagrams showing the configuration of chat fragments.
- 2A and 2B respectively show examples of the configuration of input fragments.
- 2C and 2D respectively show examples of the configuration of response fragments.
- FIG. 2A shows an example of an input fragment.
- the input fragment of FIG. 2A contains textual information. This text corresponds to the text input from the general user terminal device 7 in the chat.
- FIG. 2B shows another example of the input fragment.
- the input fragment of FIG. 2B includes reference information to feature word groups in addition to text information. That is, the input fragment may be associated with the feature word group.
- a feature word group is a single word or a group of words characterizing the input fragment.
- FIG. 2C shows an example of the response fragment. As shown, the response fragment of FIG.
- FIG. 2C contains textual information.
- This text corresponds to the text that the inference engine unit 21 of the automatic response server device 2 outputs to the general user terminal device 7 in the chat.
- FIG. 2D shows another example of the response fragment.
- the response fragment of FIG. 2D includes methods in addition to textual information. Methods are procedures (or functions) for acquiring information from the outside (for example, an external database). If the response text contains a variable, it is possible to define the method corresponding to that variable. The data obtained by the method from outside is embedded in the text as the actual value of the variable. As a specific example, when the output text is "Your account balance is ⁇ X>yen.” (Where ⁇ X> is a variable), the method corresponding to the variable ⁇ X> is a predetermined account.
- the management database is referenced, and the value of ⁇ X> is acquired using an appropriate condition (eg, the condition of the user's account number, etc.). For example, if the value of ⁇ X> is 1,000,000, the variable in the above output text is replaced and converted into the text "Your account balance is 1,000,000 yen.”
- FIG. 3 is a schematic view showing a first example of the chat pattern.
- a first example is a chat pattern consisting of one input (F1) and one response (F2).
- a feature word group is associated with the input F1.
- the automatic response server device 2 acquires the input F1 and outputs a response F2 as a response to the input F1.
- This response F2 is delivered to the general user terminal device 7.
- the text data input from the general user terminal device 7 is estimated to correspond to the input F1 according to the feature represented by the feature word group linked to the input F1.
- FIG. 4 is a schematic view showing a second example of the chat pattern.
- the second example is a chat pattern in which two or more pairs of input and response are continuous. That is, in this chat pattern, the automatic response server device 2 acquires the input F11, and outputs a response F12 as a response to the input F11. Furthermore, the automatic response server device 2 obtains the input F13, and outputs a response F14 as a response to the input F13. The following also continues.
- the inputs F11 and F13 are each associated with a feature word group.
- FIG. 5 is a schematic view showing a third example of the chat pattern.
- the third example is a pattern that starts at an input F21 and ends at a response F60, and includes a fixed pattern from an intermediate response F30.
- the fixed pattern in the middle is a series of flows in the chat fixed. That is, the feature word group is not associated with the input included in the fixed pattern. For example, since a feature word group is not associated with the input F31, it is not determined that any input text matches the input F31.
- the state of the input F31 is reached only after the automatic response server device 2 outputs the response F30.
- the input F33 is also similar to the input F31. That is, a chat performed between the automatic response server device 2 and the general user terminal device 7 does not flow in the middle of this fixed pattern.
- Such a fixed pattern can be used for the automatic response server device 2 to acquire a series of information through multiple exchanges with the general user terminal device 7. Specifically, it is possible to use such a fixed pattern in order to obtain information from the general user terminal device 7 by the automatic response server device 2 in accordance with a fixed scenario.
- a store that receives an order for a product is a series of chats including the name of the orderer, the address of the orderer, the telephone number of the orderer, the number identifying the ordered item, and the number of ordered items.
- Such fixed pattern can be used when acquiring by exchange of
- FIG. 6 is a schematic view showing a fourth example of the chat pattern.
- the fourth example includes a pattern in which a sequence from a representative input and a sequence from a similar input join together in a common response.
- the inputs F71a, F71b, and F71c (and so forth) are common in that they are inputs leading to the response F72, but they are associated with different feature word groups.
- the response system 1 of the present embodiment treats the input F71a as a representative input.
- inputs (F71 b, F71 c,...) Other than the representative input are similar inputs.
- Such chat patterns can be used to handle multiple layers of input in a single context.
- an input “I want to open an account” is used as a representative input (input F71a)
- an input “I want to create a passbook” is used as a similar input.
- a characteristic word group can be linked to each input, and both of them may be largely different from each other.
- FIG. 7 is a schematic view showing a fifth example of the chat pattern.
- the fifth example includes a pattern in which the response branches in accordance with the content of the input.
- the input F81 ⁇ the response F82 ⁇ the input F83 is continued.
- it can be branched whether to proceed to the response F84a or to proceed to the response F84b.
- a rule for input text can be described in a chat pattern.
- different responses can be described according to the conditions in the rule.
- the automatic response server device when trying to realize a service related to a time deposit account in the banking service with the automatic response server device 2, the automatic response server device depends on whether a new time deposit is set or an existing time deposit is canceled. You may want to change the 2 response. In such a case, the chat pattern shown as the fifth example can be used.
- the chat pattern may be configured by combining a plurality of patterns exemplified in FIGS. 3 to 7.
- the chat pattern may be configured to include the fixed pattern shown in FIG. 5 and to include the branches shown in FIG. The same is true for other combinations.
- the created chat pattern is expressed by appropriate data and stored in the response knowledge data storage unit 23 in the automatic response server device 2.
- the chat pattern includes a chain of inputs and responses, and may also include flow bifurcation (FIG. 7) and flow consolidation (FIG. 6).
- the input is associated with data of the feature word group.
- Data of chat patterns having such a property can be expressed, for example, in XML, but the form of data expression is not particularly limited.
- FIG. 8 is a schematic diagram showing an example of a field tree.
- the field tree has a tree structure of multiple layers (four layers in the illustrated example).
- Each node of the domain tree corresponds to the chat domain. From the viewpoint of a certain node, upper nodes (parent nodes etc.) correspond to upper concepts of the field, and lower nodes (child nodes etc.) correspond to lower concepts of the field.
- nodes in fields such as “restaurant”, “home appliances”, “cosmetics”, etc. are included in the top hierarchy (depth level 1).
- depth level 1 As a child node of "restaurant” among the nodes of depth level 1, "introduction of restaurant”, “guidance of food”, “reservation method”, “in the next hierarchy (depth level 2)" It includes nodes in the field of access, etc.
- depth level 3 As a child node of "information for cooking” among the nodes of depth level 2, in the next layer (depth level 3), "fox udon”, “tempura udon”, “super hot ramen”, “pork bone ramen” ", Etc.” contains nodes in the field.
- trees are classified according to their type of business. For each node included in the category tree, an identifier for uniquely identifying the node is given.
- the prototype database server device 4 the prototype of the chat pattern stored in the prototype storage unit 41 is stored in association with the node of the classification tree as described above.
- the member operating the member terminal 3 can trace from the classification tree when searching for a chat pattern prototype. That is, members can easily find a prototype of a desired chat pattern.
- FIG. 8 shows an example in which the form of connection between fields is a tree structure, the fields may be connected by other forms.
- field identifier an identifier (referred to as “field identifier”) to one field (corresponding to one node in FIG. 8)
- a prototype of the chat pattern stored in the prototype storage unit 41 is It becomes possible to relate to a specific field. This makes it easy for a member etc. to find a chat pattern prototype.
- FIG. 9 is a block diagram showing a schematic functional configuration of the member terminal 3.
- the member terminal 3 includes a prototype acquisition unit 31, a chat pattern editing unit 32, a response knowledge data transmission unit 33, a learning processing unit 34, and an input interpretation knowledge data transmission unit 35. And a barcode acquisition unit 36.
- the functions of these units are as follows.
- the prototype acquisition unit 31 acquires data of a chat pattern prototype from the prototype database server device 4.
- the data of the chat pattern prototype is created in advance by the operator of the prototype database server device 4.
- the user of the member terminal 3 can select data of an appropriate prototype that is consistent with his own business and download it from the prototype database server 4. For example, the user of the member terminal device 3 selects and downloads data of a prototype that matches the type of business to which the own business belongs.
- the user of the member terminal 3 can create chat pattern data by adding necessary information to the prototype downloaded from the prototype database server 4 and embedding it.
- the chat pattern editing unit 32 performs the creation of chat pattern data.
- the chat pattern editing unit 32 (chat pattern generation unit) has a function of editing a chat pattern.
- the chat pattern editing unit 32 can newly generate, correct, or partially delete chat pattern data according to the operation of the operator (member). That is, the chat pattern editing unit 32 can also create a completely new chat pattern.
- the chat pattern editing unit 32 can also create a chat pattern by modifying the prototype acquired by the prototype acquisition unit 31 from the prototype database server device 4.
- the chat pattern editing unit 32 can also change an existing (already registered) chat pattern.
- the chat pattern editing unit 32 When the chat pattern editing unit 32 generates a chat pattern based on a prototype, the field identifier associated with the original prototype is also inherited to the generated chat pattern. That is, the chat pattern may have an area identifier associated with it.
- the chat pattern editing unit 32 internally stores the chat pattern obtained as a result of the editing process.
- the chat pattern editing unit 32 passes the chat pattern obtained as a result of the editing process to the response knowledge data transmitting unit 33 and the learning processing unit 34.
- the chat pattern editing unit 32 has information of a chat pattern which is a pattern of an input fragment which is an input to the automatic response server apparatus in chat and a response fragment which is a response from the automatic response server apparatus to the input fragment. It is generated.
- the chat pattern editing unit 32 may generate information (for example, a feature word group) of the feature corresponding to the input fragment (input text).
- the response knowledge data transmitting unit 33 transmits the response knowledge data (data of response fragments) included in the chat pattern created by the chat pattern editing unit 32 to the automatic response server device 2.
- the response knowledge data transmission unit 33 transmits, to the automatic response server device 2, response knowledge data based on the response fragment included in the chat pattern generated by the chat pattern editing unit 32.
- the learning processing unit 34 performs learning processing based on the chat pattern created by the chat pattern editing unit 32, and outputs input interpretation knowledge data which is the result of the learning processing. Specifically, the learning processing unit 34 performs a learning process as follows.
- the learning processing unit 34 uses, as learning data, a set of pairs of an input text and a response (the value of RESP i in Equation (1)) corresponding to the input text.
- the learning data is provided, for example, by the member operating the member terminal 3.
- the chat pattern editing unit 32 edits the chat pattern, for example, a member may input an input text example.
- the above set of pairs is positive data for learning.
- negative example data for learning may be prepared.
- the learning processing unit 34 optimizes the parameter value group ⁇ using a method of error back propagation method using the above-described learning data.
- the learning processing unit 34 performs a learning process using a learning method suitable for the method of the inference engine unit 21 as appropriate. As described above, the learning processing unit 34 obtains and outputs knowledge data for input interpretation (for example, the parameter value group ⁇ ).
- the learning processing unit 34 based on the chat pattern generated by the chat pattern editing unit 32 and the input text example, the learning processing unit 34 relates the input text to the automatic response server device 2 and the response corresponding to the input text. Learning processing is performed and knowledge data for input interpretation obtained as a result of the learning processing is output. In other words, the learning processing unit 34 performs learning processing of the relationship between the input text corresponding to the input fragment and the response fragment based on the chat pattern generated by the chat pattern editing unit 32, and is obtained as a result of the learning processing. Output knowledge data for input interpretation.
- the input interpretation knowledge data transmission unit 35 transmits the input interpretation knowledge data obtained and output by the learning processing unit 34 to the automatic response server device 2. That is, the input interpretation knowledge data transmission unit 35 transmits the input interpretation knowledge data output from the learning processing unit 34 to the automatic response server device 2.
- the input interpretation knowledge data includes the information of the parameter value group ⁇ in the equation (1) described above.
- the response knowledge data transmission unit 33 and the input interpretation knowledge data transmission unit 35 synchronously transmit data to the automatic response server device 2.
- the response knowledge data transmitted by the response knowledge data transmitter 33 and the input interpretation knowledge data transmitted by the input interpretation knowledge data transmitter 35 are matched. That is, specifically, the learning processing unit 34 performs the learning process at the timing when the chat pattern editing unit 32 finishes the editing process of one break.
- the response knowledge data held for transmission by the response knowledge data transmission unit 33 and the input interpretation knowledge data held for transmission by the input interpretation knowledge data transmission unit 35 match each other. Also on the automatic response server device 2, the response knowledge data and the input interpretation knowledge data are respectively updated in a consistent manner.
- the barcode acquisition unit 36 (also referred to as an access information generation unit) is access information for the general user terminal device 7 to access the chat room of the member (member who operated the member terminal 3). Generate barcodes.
- the access information is typically information of a URL (Universal Resource Locator) of a service provided by the automatic response server device 2.
- the barcode acquisition unit 36 may generate a two-dimensional code or other code information as access information instead of the barcode. In any case, by using the access information generated by the barcode acquisition unit 36, the chat room of the member can be accessed.
- the barcode acquiring unit 36 generates barcode information (access information) for accessing a chat to which the result of the learning processing is reflected.
- the access information generated by the barcode acquisition unit 36 includes, for example, information for identifying a member.
- the general user terminal device 7 that has accessed the automatic response server device 2 using the access information is connected to the chat room of the member.
- the access information generated by the barcode acquisition unit 36 includes, as necessary, information of a field identifier.
- the general user terminal device 7 that has accessed the automatic response server device 2 using the access information is connected to the chat pattern of the member who matches the field identifier.
- the access information generated by the barcode acquisition unit 36 includes information for identifying a specific chat pattern.
- the general user terminal device 7 that has accessed the automatic response server device 2 using the access information is connected to the chat pattern of the member who matches the identification information.
- the barcode acquisition unit 36 may generate the access information only by its own processing in accordance with a predetermined rule.
- the barcode acquisition unit 36 may generate the access information in the form of acquiring at least a part of the access information from the automatic response server device 2.
- the barcode acquisition unit 36 may acquire from the automatic response server device 2 the access information itself or all the information required to generate the access information.
- the barcode acquisition unit 36 may acquire the completed barcode image data itself from the automatic response server device 2 and output it.
- the automatic response server device 2 provides the barcode acquisition unit 36 with necessary information as appropriate.
- the user of the member terminal device 3 appropriately prints the access information output from the bar code acquisition unit 36 on a paper medium or the like so that the customer or the like of the own business can access the chat.
- the user of member terminal device 3 displays, for example, access information (bar code, two-dimensional code, etc.) printed on a paper medium in the store or, in the case of a restaurant, it is included in a menu such as cooking Or distributed as a distribution.
- knowledge data and chat patterns are updated as follows. That is, the member operating the member terminal 3 corrects the existing chat pattern based on the notice etc. obtained through the daily work. Alternatively, a member operating the member terminal 3 generates and registers a new chat pattern in order to provide a new service as a business. The chat pattern updated in this way is reflected to the automatic response server device 2 at a predetermined timing.
- FIG. 10 is a schematic view showing an example of a screen for registering a chat pattern in the member terminal 3. As shown in FIG. As shown, this screen is implemented using a graphical user interface. As illustrated, this chat pattern registration screen has fields for inputting a title, a representative input, and a plurality of similar inputs. This chat pattern registration screen has an "increase” button and three buttons for selecting a category. In the illustrated example, there is a "Shinkansen” button as category 1, a "subway” button as category 2, and a "bus” button as category 3. The chat pattern registration screen has a field for inputting a member response. The chat pattern registration screen has a “download” button, a “register” button, a “delete” button, and a “save” button.
- chat pattern registration screen shown in FIG. 10 is suitable for editing the chat pattern shown in FIG. 6 (one representative input sequence and a plurality of similar input sequences correspond to one response).
- a screen for setting an input fragment and a response fragment is appropriately prepared.
- the title is a title given to each chat pattern.
- An arbitrary character string can be used as the title.
- the representative input corresponds to the representative input (see FIG. 6) among the inputs included in this chat pattern.
- the similar input corresponds to the similar input (see FIG. 6) corresponding to the representative input among the inputs included in the chat pattern.
- three similar inputs can be written. By pressing the "increase" button, the number of fields for writing similar inputs can be increased.
- Each button of Category 1 (“Shinkansen”), Category 2 (“Metro”), and Category 3 (“Bus”) is a column for specifying the category to which this chat pattern belongs.
- Division 1, Division 2, and Division 3 each have a predetermined depth level in the field tree shown in FIG. 8 (Depth Level 1, Depth Level 2, Depth Level 3, or Depth Level 4) It corresponds to the node in.
- the user of the member terminal device 3 can edit the chat pattern belonging to the node by pressing any one of the group supplements on the chat pattern registration screen. That is, the chat pattern registered from the chat pattern registration screen is registered as the chat pattern belonging to the node (field) selected by pressing the button. That is, the chat pattern to be registered can be associated with a specific field identifier.
- the member response column is a column for setting the text of the response corresponding to the representative input or the similar input.
- the “register” button is a button for performing processing of registering the chat pattern created on this screen.
- the chat pattern is passed from the chat pattern editing unit 32 to the response knowledge data transmitting unit 33 and the learning processing unit 34.
- the response knowledge data transmission unit 33 sequentially receives one or more response knowledge data from the chat pattern editing unit 32, and when there is a separately provided trigger, the response knowledge data is transmitted to the automatic response server device 2 Send to
- the learning processing unit 34 sequentially receives a chat pattern or a plurality of chat patterns from the chat pattern editing unit 32, and executes a learning process based on the chat patterns when there is a separately provided trigger.
- the “delete” button is a button for performing processing for deleting the chat pattern opened on this screen.
- the chat pattern opened on this screen is deleted in the member terminal 3.
- Information indicating that the chat pattern has been deleted is passed from the chat pattern editing unit 32 to the response knowledge data transmitting unit 33 and the learning processing unit 34.
- the response knowledge data transmitting unit 33 receives information on the deleted chat pattern from the chat pattern editing unit 32. Then, the response knowledge data transmitting unit 33 automatically receives the information that the response knowledge data included in the deleted chat pattern is to be deleted when there is a separately provided trigger.
- Send to The learning processing unit 34 receives information on the deleted chat pattern from the chat pattern editing unit 32. Then, when there is a separately provided trigger, the learning processing unit 34 executes learning processing again in a state where there is no deleted chat pattern.
- the “save” button is a button for performing processing for temporarily saving the chat pattern edited on this screen.
- the chat pattern opened on this screen is saved in the storage means in the member terminal 3.
- the response fragment information included in the chat pattern that is, the response knowledge data is not sent to the automatic response server device 2 only by being saved by the “save” button. If stored only by the "save” button, learning processing based on the chat pattern is not performed.
- the user of the member terminal 3 can call up and re-edit the chat pattern saved by the "Save” button later.
- the user of the member terminal 3 can register the re-edited chat pattern.
- the user of the member terminal 3 can also delete the chat pattern once stored.
- the “download” button is a button for performing processing of downloading a prototype of a chat pattern that satisfies a specific condition from the prototype database server device 4.
- the user of the member terminal device 3 can edit, register or save the chat pattern using the chat pattern prototype downloaded from the prototype database server device 4.
- a feature word group corresponding to an input fragment in the chat pattern is automatically extracted from the text of each input (representative input or similar input).
- the user of the member terminal 3 may manually set the feature word group corresponding to the input fragment using a screen (not shown).
- the user of member terminal device 3 designates a division appropriately and presses the "download” button to download a chat pattern prototype from prototype database server device 4, and further The text is entered in the member response column.
- the case where the user of the member terminal device 3 selects the “Shinkansen” of Category 1 is shown.
- a prototype of a chat pattern titled “Go” is obtained from the prototype database server device 4.
- the text "Please tell me the way” is set as a representative input.
- the texts “How can I do it?”, “How can I get to the store?”, And “Tell me directions for transportation” are set.
- the texts of these representative inputs and similar inputs are written in advance in the prototype database server device 4 by the operator of the prototype database server device 4 as prototype data.
- the user of the member terminal 3 can use the texts of these representative inputs and similar inputs as they are, or can edit and change them.
- the user of member terminal device 3 has set the text of the response appropriate for his own business in the member response column in response to the input (representative input or similar input) (keyboard etc. Typed in from the character input means of Specifically, in the column of member response, by the user of member terminal device 3, "Please exit from the Yaesu central entrance of Shinkansen Tokyo Station. Exit the exit and walk about 100 m to the left. Kiosk Walk 30 meters to the right from where you can see the convenience store. The 3rd floor of the building will be the store. " That is, the user of the member terminal 3 can set the text of the response suitable for his / her business according to the contents (representative input and similar input) of the prototype created by the operator of the prototype database server device 4 .
- the user of the member terminal 3 can open the screen of FIG. 10 to set the input fragment and the response fragment of the chat pattern. Associating input fragments with response fragments in one chat pattern is also performed on this screen. As described above, the user of the member terminal 3 can register the various patterns shown in FIGS. 3 to 7 by appropriately associating the input fragment with the response fragment.
- FIG. 11 is a schematic view showing an example of a screen for editing and registering the synonym dictionary in the member terminal device 3.
- the synonym dictionary registration screen has a field for entering a search term and a "search" button.
- the synonym dictionary registration screen has a field for inputting a combination of additional keywords and synonyms, and a plurality of “register” buttons and “delete” buttons. In the illustrated example, three sets are displayed, but the number may be arbitrarily changed.
- fields for adding keywords a field for adding keywords and a field for synonyms thereof
- a “register” button, a “delete” button, and a “save” button are provided for keyword addition.
- the search term column is a section for setting a search term for searching for a keyword.
- the “search” button is a button for giving a trigger for executing a keyword search using the search term. When this search is executed, a list of corresponding keywords and their synonyms is displayed and can be edited. In the example of the state of the screen to illustrate, "business hours" is set as a search term.
- the keyword column is a column for inputting a key word for synonym registration.
- the synonym column is a column for entering one or more synonyms associated with a keyword. That is, in the first line of the illustrated example, the keywords “opening time” correspond to synonyms such as “starting time”, “opening time”, and “opening time”. In the synonym section, it is possible to set multiple synonyms separated by slashes. In the second line, the keywords “business start” correspond to synonyms "business start time”, “business start time”, and “business start”. In the third line, the keywords "last order” correspond to synonyms such as “last order”, "final order”, and "final order”. For example, it may be possible to interpret typographical errors (incorrect input) in chat text as words according to their original intentions by registering keyword homophones as synonyms.
- Weight values may be set for keyword and synonym pairs.
- the weight is a numerical value representing the weight of the set.
- the weight value is a real number of 0 or more. This weighting value represents the degree to which the synonyms in the set are used as synonyms of the keyword. The higher the weight value is, the higher the weight value is used in place of the keyword. Conversely, the lower the weight value, the lower the weight value is used in place of the keyword.
- the “register” button corresponding to the keyword (symbol) pair (row) is a button serving as a trigger for registering the keyword and the synonym entered in the row in the synonym dictionary.
- the "delete” button is a button serving as a trigger for deleting the synonym corresponding to the keyword from the synonym dictionary.
- Synonyms registered in the synonym dictionary have the possibility of being replaced with keywords when the inference engine unit 21 processes input text in chat. Whether or not the inference engine unit 21 actually uses the synonym also depends on the registered weight value. Feature words may be expanded (replaced) by using a synonym dictionary. By permitting substitution using a synonym dictionary, it is possible to absorb the fluctuation of expression in the input text and interpret the text. When synonym substitution is performed, it is also possible to control substitution easiness by setting the above-mentioned weight value appropriately.
- FIG. 12 is a schematic view showing an example of implementation of chat performed between the general user terminal device 7 and the virtual user in the automatic response server device 2.
- the figure (a) shows an example of the external appearance (plan view) of the general user terminal device 7.
- the figure (b) shows an example of the external appearance (oblique view) of the object (fruit) which printed the code information (access information) which the barcode acquisition part 36 of the member terminal device 3 acquired.
- the printed code information (access information) may be a bar code, a two-dimensional code, characters, and other forms of information.
- the printed code information is, for example, optically readable.
- the figure (c) is an example of a screen of a chat room displayed on the general user terminal unit 7 side.
- the procedure for accessing the chat service from the general user terminal device 7 will be described with reference to FIG. (1)
- the member terminal device 3 (not shown) performs a learning process based on the set chat pattern.
- the member terminal 3 transmits the response knowledge data and the input interpretation knowledge data to the automatic response server device 2.
- the automatic response server device 2 stores the response knowledge data and the input interpretation knowledge data transmitted from the member terminal 3 side.
- the automatic response server device 2 is a member who has access information for accessing the chat service of the member, or access information for accessing the chat service associated with a specific area identifier of the member. Transmit to terminal device 3. For example, whether or not to access the chat service associated with a specific field identifier in the member can be set as appropriate, for example.
- the member terminal device 3 Based on the access information received from the automatic response server device 2, the member terminal device 3 prints out an optically readable two-dimensional code.
- the two-dimensional code shown in FIG. 12 (b) is an example.
- the object on which the two-dimensional code is printed is placed, for example, in a store or the like.
- the user of the general user terminal device 7 has the general user terminal device 7 (specifically, for example, a smartphone, a tablet terminal, etc.) shown in FIG. This two-dimensional code is read by holding it over the two-dimensional code.
- the general user terminal device 7 extracts information for accessing the chat service from the read two-dimensional code. Specifically, the general user terminal device 7 extracts, for example, URL information for accessing a service of a specific chat room.
- the general user terminal device 7 uses the communication means to access the server device indicated by the URL.
- the URL is information indicating the location of a specific service in chat server device 5.
- the general user terminal device 7 can access a chat room corresponding to a member store or corresponding to specific information in the member store.
- chat room screen The title of the window on the screen shown in FIG. 12C is displayed as "chat room screen".
- the right side (the user is described as "Naoko Toda") is the user of the general user terminal device 7, and the left side (the user is described as the "shop owner") is the automatic response server. It is a virtual user in the device 2.
- the user of the general user terminal device 7 on the right side inputs "Do you have a children's menu?"
- the inference engine unit 21 of the automatic response server device 2 estimates an optimal response fragment based on the input text. At the time of this estimation, parameter values obtained in advance by learning processing are used.
- the parameter values are stored in the input interpretation knowledge data storage unit 22 of the automatic response server device 2.
- the inference engine unit 21 of the automatic response server device 2 returns a response according to the response fragment.
- the automatic response server device 2 returns the text of the response, "Yes! Of course. The menu for children is also satisfactory. Come with your children.”
- the text of this response is delivered to the general user terminal device 7 via the chat server device 5 and displayed on the screen.
- the inference engine unit performs the same process as described above for the text "Yes? You are saved. Thank you.” As a result, it outputs the text of the response "No, welcome, please take care.”
- Another example (not shown) in the chat service is as follows.
- the inference engine unit 21 estimates the optimal response fragment, and as a result, "Thank you for your inquiry.
- the balance of the account is 10,000,000 yen.
- the automatic response server device 2 returns the text of the response.
- the balance amount "10,000,000" is a value of a variable included in the response fragment in the corresponding chat pattern.
- the method specified in the response fragment refers to the account information database and obtains the balance of the bank account of the user of the general user terminal device 7. The method substitutes the acquired value of the balance as the value of the above variable included in the response fragment.
- step 1 a member registers as a member from the homepage of the operator of the automatic response server apparatus 2 (hereinafter referred to as A company).
- a company the operator of the automatic response server apparatus 2
- the prototype database server device 4 of company A can be accessed.
- the member uses the member terminal device 3 to download a prototype corresponding to himself from the tree (field tree) according to each industry field from the prototype database server device 4.
- step 2 the member writes a response (answer to inform the general user) along the own business in accordance with the downloaded representative input (representative question) and similar input (similar question). Thus, the member completes the chat pattern.
- Step 3 the member uses the member terminal 3 to execute a learning process (machine learning process).
- the following two files are obtained as a result of the learning process.
- Input interpretation knowledge data Data including knowledge (parameters) representing the relationship between input text and response.
- Knowledge data for response Data obtained by a member inputting a response fragment (answer) to the representative input and the similar input.
- step 4 from the member terminal device 3, the above-described two data generated by the learning process and the like are input to the input interpretation knowledge data acquisition unit 24 and the response knowledge data acquisition unit 25 in the automatic response server device. Transmit each one.
- step 5 the automatic response server device 2 stores the input interpretation knowledge data and the response knowledge data received from the member terminal device 3 in the respective storage units.
- the automatic response server device 2 transmits, to the member terminal device 3, URL information (access information) for directly accessing the chat room associated with the data transmitted from the member terminal device 3 side.
- the access information is, for example, a barcode or two-dimensional code information.
- the member terminal 3 generates and outputs this access information on the terminal side.
- step 7 the member prints out the access information (bar code and the like) on a shop menu or a table where the general user can see well.
- a general user can easily hold a bar code on the general user terminal device 7 at any time to connect to the related chat room.
- the general user terminal device 7 connects to the chat room and transmits the input text.
- the general user terminal device 7 transmits the text "What can I do?"
- the inference engine unit 21 of the automatic response server device 2 is based on the input interpretation knowledge data based on the features of the received input text (for example, the feature words included in the input text are the features of the input text). , To estimate the optimal response.
- the input text "What can I do?" Has features similar to those of "How can I do it", which is one of the similar inputs in FIG.
- the inference engine unit 21 generates and outputs a response text by referring to the response knowledge data based on the response (response fragment) corresponding to the direction. According to this flow of the above processing, general users can ask what they want to know and obtain necessary answers through processing by artificial intelligence.
- step 9 among the inputs (questions) issued by the general user, there are cases where the inference engine unit 21 can not answer (for example, the likelihood of the estimated optimal response is lower than a predetermined threshold)
- the automatic response server device 2 notifies the member terminal device 3 as an alarm.
- step 10 the member terminal device 3 resets the chat pattern from the chat pattern registration screen with respect to the input text for which the alarm has been received, and re-executes the learning processing as appropriate.
- the result of the re-executed learning process is stored as knowledge data on the side of the automatic response server device 2 by the same process as described above. That is, the learning process can be repeated as appropriate to update the knowledge data.
- the member terminal device 3 can download the data serving as the prototype of the chat pattern from the prototype database server device 4, so that it is possible to easily launch a response system suitable for the type of business of its own business.
- a chat pattern is registered in the member terminal device 3, and knowledge data is generated by learning processing in the member terminal device 3 based on the chat pattern.
- Those chat patterns and knowledge data are sent to the automatic response server device 2.
- the automatic response server device 2 updates the input interpretation knowledge data storage unit 22 and the response knowledge data storage unit 23 using data (input interpretation knowledge data and response knowledge data) received from the member terminal 3 Do.
- the learning process is performed on the member terminal 3 side. That is, the automatic response server device 2 may not perform the learning process for generating the knowledge data used by itself.
- the member terminal device 3 can be realized by, for example, a PC, a tablet terminal, a smartphone or the like.
- the automatic response server device 2 can be constructed relatively inexpensively.
- the calculation of the learning process for that amount can be performed using the remaining capacity of the member terminal 3 held by the members. As a result of them, the entire response system 1 can be constructed at low cost comprehensively.
- the member terminal 3 outputs access information (for example, a barcode, a two-dimensional code, etc.) for accessing a service of a specific chat room.
- access information for example, a barcode, a two-dimensional code, etc.
- the general user terminal device 7 reads this access information using, for example, an optical reader (reading means)
- the general user terminal device 7 can obtain the service of the desired chat room (in other words, the desired member's membership It is possible to easily access the service of the chat room of the person or the chat room of the field of the specific information provided by the desired member (for example identified by the field identifier).
- FIG. 13 is a block diagram showing a schematic functional configuration of a response system according to the second embodiment.
- the response system 101 includes an automatic response server device 102, a member terminal device 3, a prototype database server device 4, a chat server device 5, and a general user terminal device 7. .
- the feature of this embodiment is that it includes an automatic response server device 102 in place of the automatic response server device 2 in the previous embodiment.
- the automatic response server device 102 includes an inference engine unit 21, an input interpretation knowledge data storage unit 22, a response knowledge data storage unit 23, an input interpretation knowledge data acquisition unit 24, and response knowledge. It comprises a data acquisition unit 25 and an additional learning processing unit 128.
- the functions of the inference engine unit 21, the input interpretation knowledge data storage unit 22, the response knowledge data storage unit 23, the input interpretation knowledge data acquisition unit 24, and the response knowledge data acquisition unit 25 are the same. , The same as those in the first embodiment.
- the feature of the automatic response server apparatus 102 in the present embodiment is that it has an additional learning processing unit 128.
- the additional learning processing unit 128 performs additional learning processing based on the text of the actual chat between the virtual user processed by the inference engine unit 21 and the user of the general user terminal device 7. Specifically, the additional learning processing unit 128 acquires the text of the actual chat. The additional learning processing unit 128 externally acquires data representing whether or not the determination (the estimation of the corresponding input fragment) of the input text by the inference engine unit 21 in the acquired chat is a correct answer. For example, a person judges and inputs whether or not the answer is correct. Then, the additional learning processing unit 128 performs learning processing with the chat determined as the correct answer as a positive example and the chat determined as an error (not the correct answer) as a negative example.
- the learning process itself using the positive example and the negative example is the same as the learning process described as the process by the learning processing unit 34 in the first embodiment. That is, an error back propagation method is used as an example.
- the additional learning processing unit 128 passes the knowledge data, which is the result of the learning processing, to the input interpretation knowledge data acquiring unit 24. Then, the input interpretation knowledge data acquisition unit 24 updates the input interpretation knowledge data storage unit 22 using the acquired knowledge data.
- the learning process is performed based on the text of the actual chat, the accuracy of the knowledge data is further improved. That is, it is possible to further improve the quality of chat by the automatic response server apparatus 102.
- FIG. 14 is a block diagram showing a schematic functional configuration of a response system according to the third embodiment.
- the response system 201 includes an automatic response server device 202, a member terminal device 3, a prototype database server device 4, a chat server device 5, and a general user terminal device 7. .
- the feature of this embodiment is that it includes an automatic response server device 202 in place of the automatic response server device 102 in the previous embodiment.
- the automatic response server device 202 includes an inference engine unit 21, an input interpretation knowledge data storage unit 22, a response knowledge data storage unit 23, an input interpretation knowledge data acquisition unit 24, and response knowledge.
- a data acquisition unit 25, an additional learning processing unit 128, and a barcode generation unit 229 (access information generation unit) are included.
- the functions of the inference engine unit 21, the input interpretation knowledge data storage unit 22, the response knowledge data storage unit 23, the input interpretation knowledge data acquisition unit 24, and the response knowledge data acquisition unit 25 are the same.
- the function of the additional learning processing unit 128 is the same as that in the second embodiment.
- the feature of the automatic response server device 202 in the present embodiment is that it has a barcode generator 229.
- the barcode generation unit 229 generates optically readable code information including information (access information) for accessing a service of a specific chat room. For example, the barcode generation unit 229 generates barcode, two-dimensional code, code information of a character string readable by an OCR (optical character reader), and the like. Then, the barcode generation unit 229 transmits the generated code information to the member terminal device 3. Specifically, for example, the barcode generation unit 229 transmits image data including a barcode, a two-dimensional code, and the like to the member terminal 3.
- the barcode acquisition unit 36 on the member terminal 3 side receives the optically readable code information transmitted from the barcode generation unit 229, and prints out.
- the general user terminal device 7 can optically read the optically readable code information thus output.
- the automatic response server device generates code information that can be read optically (that is, image information that can read access information optically). As a result, it becomes possible to centrally generate and manage optically readable code information on the automatic response server device side.
- the feature word extracted from the text was used as a feature of the text inputted in chat.
- Feature words may be expanded (replaced) by using a synonym dictionary.
- another feature is used as a feature of the input text.
- syntactic analysis or dependency analysis of input text is performed, and a tree (parsing tree or dependency analysis tree) which is an analysis result is used as a feature.
- using the result of syntactic analysis or dependency analysis as a feature may improve the accuracy of grasping the content of the input text as compared with the case of simply extracting the feature word.
- the two feature words co-occur or occur far in the parsing tree or the dependency parsing tree. It can be treated as a different feature depending on whether it co-occurs with.
- “close” or “far” can be achieved, for example, by the distance between the nodes of the tree (the number of hops of the arc connecting the node and the node).
- the automatic response server device 2 (or 102), the prototype database server device 4 and the chat server device 5 are configured as independent devices (such as computers). In the second modification, a plurality of these server devices are integrated, and the plurality of functions are implemented in one device.
- the automatic response server device 2 (102) and the prototype database server device 4 are integrated into one device.
- the automatic response server device 2 (102) and the chat server device 5 are integrated into one device.
- the prototype database server device 4 and the chat server device 5 are integrated into one device.
- the automatic response server device 2 (102), the prototype database server device 4 and the chat server device 5 may be integrated into one device.
- the synonym dictionary may be registered.
- existing synonym dictionary data is prepared in advance, and the inference engine unit 21 can refer to the synonym dictionary data.
- the functions (or part of the functions) of the devices in the above-described embodiment and modification may be realized by a computer.
- a program for realizing this function may be recorded in a computer readable recording medium, and the program recorded in the recording medium may be read and executed by a computer system.
- the "computer system” mentioned here includes hardware such as an OS and peripheral devices.
- Computer-readable recording medium means portable media such as flexible disks, magneto-optical disks, ROMs, CD-ROMs, DVD-ROMs, USB memories, etc., and storage devices such as hard disks built in computer systems. Say.
- “computer-readable recording medium” holds a program dynamically for a short time, like a communication line in the case of transmitting a program via a network such as the Internet or a communication line such as a telephone line. It may also include one that holds the program for a certain period of time, such as volatile memory in the computer system that becomes the server or client in that case.
- the program may be for realizing a part of the functions described above, or may be realized in combination with the program already recorded in the computer system.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computer Networks & Wireless Communication (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Information Transfer Between Computers (AREA)
- Machine Translation (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Computer And Data Communications (AREA)
Abstract
Description
スムーズで的確なチャットを実現するために、事前に行っておく学習処理が重要である。特に、頻繁に学習処理を行うことが求められる適用分野では、入出力の実行の処理だけではなく、学習の処理にも充分な計算資源(CPU時間等)を割り当てる必要がある。
しかしながら、チャットの入出力の実行の処理と学習の処理との両方を充分に行うためには、大規模で高価な計算機を用いる必要がある。特に、学習処理は、多大な計算資源を消費するものである。つまり、構築すべきシステムが高コストなものになるという問題があった。
特に、本発明の一態様によれば、加盟者端末装置の使用者は、プロトタイプデータベースサーバー装置から、事前に準備されたチャットパターンのプロトタイプをダウンロードする。そして、加盟者端末装置の使用者は、そのプロトタイプに基づいて、入力(例えば、質問)に対する応答(例えば、その質問に対する回答)を設定するなどして、チャットパターンを作成する。これにより、より一層手軽に、自事業用にカスタマイズされた応答システムを構築することが可能となる。ここで、チャットパターンのプロトタイプとは、例えば、代表入力や類似入力の例を含むものである。
図1は、第1実施形態による応答システムの概略機能構成を示すブロック図である。図示するように、応答システム1は、自動応答サーバー装置2と、加盟者端末装置3(端末装置)と、プロトタイプデータベースサーバー装置4と、チャットサーバー装置5と、一般利用者端末装置7とを含んで構成される。自動応答サーバー装置2と、加盟者端末装置3と、プロトタイプデータベースサーバー装置4と、チャットサーバー装置5とは、有線または無線の通信回線等を介して相互に通信可能である。この通信においては、例えば、インターネットプロトコル(Internet Protocol)が用いられる。
自動応答サーバーのサービスを利用する加盟者は、加盟者端末装置3を保持し、利用する。加盟者端末装置3において稼働するソフトウェアの一部を、自動応答サーバー運営事業者が提供するようにしてもよい。
チャットサービス運営事業者は、チャットサーバー装置5を保持し、運用する。
チャットサービス運営事業者が上記の自動応答サーバー運営事業者と同一であってもよい。
一般利用者は、一般利用者端末装置7を用いてチャットサーバー装置5にアクセスし、サービスを受ける。一般利用者は、不特定の多数であってもよい。
自動応答サーバー装置2は、チャットサービスを介して入力されるデータを受け取り、その入力に対して適切な応答を返す機能を実装した装置である。自動応答サーバー装置2は、複数の加盟者のためのサーバーとして運営される。加盟者とは、例えば、店舗や、レストランや、金融機関や、その他の事業者等である。
応答システム1においては特に、チャットサーバー装置5は、一般利用者端末装置7と、自動応答サーバー装置2によって実行される仮想利用者との間のチャットの実施を可能とする。チャットサーバー装置5を用いたチャットは、テキスト以外のタイプのデータ(マルチモーダルデータ)を用いて行うこともできるが、ここでは、テキストのデータのみを用いて行われるチャットについて説明する。一般的には、チャットサーバー装置5は、利用者Aと利用者Bとの間のチャットを実現する。利用者Aによって入力されたテキストデータと、利用者Bによって入力されたテキストデータは、両利用者が共有するチャットルーム(仮想的な部屋)において表示される。利用者Aおよび利用者Bは、このチャットルームで交わされるテキストによる対話を画面等に表示させることができる。応答システム1においては、チャットを利用者の一方が、自動応答サーバー装置2上の仮想的な利用者である。具体的には、自動応答サーバー装置2のプログラムは、API(アプリケーション・プログラム・インターフェース)を介して、チャットルーム内のテキストデータを取得したり、チャットルームにテキストデータを送信したりする。これにより、一般利用者端末装置7(利用者A)と自動応答サーバー装置上の仮想的な利用者(利用者B,プログラム)との間のチャットを行うことができる。
このように、チャットの記録を残すことにより、自動応答サーバー装置2は、データベースに格納されたデータを、ビッグデータとして、分析対象として活用することができるようになる。
言い換えれば、入力解釈用知識データ記憶部22が記憶する知識データとは、一般利用者端末装置7から渡される入力テキストに対応して、応答断片の集合内の最適な応答断片を推定するためのパラメーター値群である。
入力解釈用知識データ記憶部22は、加盟者端末装置3の入力解釈用知識データ送信部35から送信された入力解釈用知識データを記憶する。
応答用知識データ記憶部23は、加盟者端末装置3の応答用知識データ送信部33から送信された応答用知識データを記憶するものである。
知識データ(入力解釈用知識データ、および応答用知識データ)の元となるチャットパターンの具体例については、後で、図2から図7までを参照しながら説明する。
応答用知識データ取得部25は、加盟者端末装置3から、応答用知識データを取得し、その応答用知識データを応答用知識データ記憶部23に書き込む。
加盟者端末装置3側における処理等の詳細と、知識データおよびチャットパターンの更新(書き込み)のタイミング等については、後で説明する。
分野ツリー管理部42(分野情報管理部)は、分野ツリーのデータを記憶し、管理する。分野ツリーは、分野を表すラベルと、その分野を一意に識別するための識別子とを含む。分野ツリーは、ツリー構造で分野を管理する。分野ツリーの実例については、後で図面を参照しながら説明する。
つまり、分野ツリー管理部42は、プロトタイプが属する分野の情報を記憶し、管理する。
チャットパターンは適切な形式のデータとして表現可能である。一例として、チャットパターンは、XML(拡張マークアップ言語,extensible markup language)で表現可能である。
図2Aは、入力断片の一例を示す。図示するように、図2Aの入力断片は、テキストの情報を含む。このテキストは、チャットにおいて一般利用者端末装置7側から入力されるテキストに対応するものである。
図2Bは、入力断片の他の例を示す。図示するように、図2Bの入力断片は、テキストの情報に加えて、特徴語群への参照情報を含む。つまり、入力断片は、特徴語群に関連付けられていてよい。特徴語群は、その入力断片を特徴づける単数の語、または複数の語の群である。
図2Cは、応答断片の例を示す。図示するように、図2Cの応答断片はテキストの情報を含む。このテキストは、チャットにおいて、自動応答サーバー装置2の推論エンジン部21が一般利用者端末装置7に向けて出力するテキストに対応するものである。
図2Dは、応答断片の他の例を示す。図示するように、図2Dの応答断片は、テキストの情報に加えて、メソッドを含む。メソッドは、外部(例えば外部のデータベース)から情報を取得するための手続き(あるいは、関数)である。応答のテキスト内に変数を含む場合に、その変数に対応するメソッドを定義することが可能である。メソッドが外部から取得したデータが、変数の実際の値としてテキスト内に埋め込まれる。具体的な一例として、出力テキストが「あなたの口座の残高は<X>円です。」(ここで、<X>は変数)であるとき、変数<X>に対応するメソッドが、所定の口座管理データベースを参照し、適切な条件(例えば、ユーザーの口座番号の条件等)を用いて<X>の値を取得する。例えば<X>の値が1,000,000である場合、上記の出力テキスト内の変数が置換され、「あなたの口座の残高は1,000,000円です。」というテキストに変換される。
図3は、チャットパターンの第1例を示す概略図である。第1例は、1個の入力(F1)と、1個の応答(F2)とから成るチャットパターンである。入力F1には、特徴語群が関連付けられている。この例では、一般利用者端末装置7から入力F1に相当する入力があったとき、自動応答サーバー装置2は、入力F1を取得し、入力F1への応答として、応答F2を出力する。この応答F2は、一般利用者端末装置7に届けられる。
一般利用者端末装置7から入力されるテキストデータは、入力F1に紐付けられている特徴語群が表す特徴によって、入力F1に当たるものであると推定される。
これらの入力の中で、代表入力以外の入力(F71b,F71c,・・・)は類似入力である。このようなチャットパターンは、ある一つのコンテキストにおいて、入力における複数の表層を扱うために利用することができる。一例として、銀行業のサービスにおいて普通預金口座を新たに開設するためのやりとりを自動応答サーバー装置2に行わせようとした場合、次の2通りの入力テキストを、それぞれ、代表入力および類似入力として扱うことができる。即ち、「口座を開きたい」という入力を代表入力(入力F71a)とし、「通帳を作りたい」(入力F71b)という入力を類似入力とする。それぞれの入力に対して特徴語群を紐付けることができ、それら両者が互いに大きく異なっていてもよい。
図8は、分野ツリーの例を示す概略図である。図示するように、分野ツリーは、多階層(図示する例では4階層)のツリー構造を有する。分野ツリーの各ノードが、チャットの分野に対応するものである。あるノードから見て、上位のノード(親ノード等)は分野の上位概念に対応し、下位のノード(子ノード等)は分野の下位概念に対応する。
分野ツリーに含まれる各ノードについて、ノードを一意に識別するための識別子が付与されている。
プロトタイプデータベースサーバー装置4において、プロトタイプ記憶部41に記憶されているチャットパターンのプロトタイプを、上記のような分類ツリーのノードに関連付けて記憶させるようにする。これにより、加盟者端末装置3を操作する加盟者は、チャットパターンのプロトタイプを探す際に、分類ツリーからたどることが可能となる。つまり、加盟者は、所望のチャットパターンのプロトタイプを容易に見つけやすくなる。
図9は、加盟者端末装置3の概略機能構成を示すブロック図である。図示するように、加盟者端末装置3は、プロトタイプ取得部31と、チャットパターン編集部32と、応答用知識データ送信部33と、学習処理部34と、入力解釈用知識データ送信部35と、バーコード取得部36と、を含んで構成される。これら各部の機能は次の通りである。
例えば、加盟者端末装置3の使用者は、自事業が属する業種に合ったプロトタイプのデータを選んでダウンロードする。加盟者端末装置3の使用者は、プロトタイプデータベースサーバー装置4からダウンロードされたプロトタイプに、必要な情報を追加して埋め込むことで、チャットパターンのデータを作成できる。チャットパターンのデータの作成は、次のチャットパターン編集部32が行う。
チャットパターン編集部32がプロトタイプに基づいてチャットパターンを生成した場合、元のプロトタイプに関連付けられていた分野識別子は、生成されたチャットパターンにも引き継がれる。つまり、チャットパターンには、分野識別子が関連付けられている場合がある。
したがって、学習処理部34は、学習用データとして、入力テキストと、その入力テキストに対応する応答(式(1)におけるRESPiの値)のペアの集合を用いる。この学習用データは、例えば、加盟者端末装置3を操作する加盟者によって与えられる。チャットパターン編集部32においてチャットパターンを編集する際に、例えば加盟者が入力テキスト例を入力するようにしてもよい。上記のペアの集合は、学習のための正例のデータである。
適宜、学習のための負例のデータを準備してもよい。
推論エンジン部21がニューラルネットワークを用いる場合、学習処理部34は、上記の学習用データを用いて、誤差逆伝搬法(backpropagation)の手法を用いて、パラメーター値群θを最適化する。推論エンジン部21がニューラルネットワーク以外を用いる場合には、適宜、推論エンジン部21の手法に合った学習手法を用いて、学習処理部34は、学習処理を行う。
以上により、学習処理部34は、入力解釈用知識データ(例えば、パラメーター値群θ)を得て、出力する。
自動応答サーバー装置2上においても、応答用知識データと入力解釈用知識データとは整合する形でそれぞれ更新される。
特に、バーコード取得部36は、学習処理部34による学習処理が完了すると、当該学習処理の結果が反映されるチャットにアクセスするためのバーコード情報等(アクセス情報)を生成する。
バーコード取得部36が生成するアクセス情報は、必要に応じて、分野識別子の情報を含む。このとき、そのアクセス情報を用いて自動応答サーバー装置2にアクセスした一般利用者端末装置7は、当該加盟者における、その分野識別子にマッチするチャットパターンに接続される。
バーコード取得部36が生成するアクセス情報は、特定のチャットパターンを識別する情報を含む。このとき、そのアクセス情報を用いて自動応答サーバー装置2にアクセスした一般利用者端末装置7は、当該加盟者における、その識別情報にマッチするチャットパターンに接続される。
図10は、加盟者端末装置3において、チャットパターンを登録するための画面の例を示す概略図である。
図示するように、この画面は、グラフィカル・ユーザー・インターフェースを用いて実現されている。図示するように、このチャットパターン登録画面は、タイトルと、代表入力と、複数の類似入力とを入力するための欄を有している。このチャットパターン登録画面は、「増やす」ボタンと、区分を選択するための3つのボタンを有している。図示する例では、区分1として「新幹線」ボタン、区分2として「地下鉄」ボタン、区分3として「バス」ボタンが存在する。このチャットパターン登録画面は、加盟者応答を入力するための欄を有している。このチャットパターン登録画面は、「ダウンロード」ボタンと、「登録」ボタンと、「削除」ボタンと、「保存」ボタンとを有している。
代表入力は、このチャットパターンに含まれる入力のうちの代表入力(図6を参照)にあたるものである。
類似入力は、このチャットパターンに含まれる入力のうち、上記の代表入力に対応する類似入力(図6を参照)にあたるものである。図示する例では、3種類の類似入力を書き込むことができる。「増やす」ボタンを押下することにより、類似入力を書き込むための欄を増やすことができる。
「保存」ボタンによって保存されたチャットパターンを、加盟者端末装置3の使用者は、後で呼び出して再編集することができる。加盟者端末装置3の使用者は、再編集したチャットパターンを登録することができる。加盟者端末装置3の使用者は、一旦保存されたチャットパターンを削除することもできる。
あるいは、不図示の画面を用いて、入力断片に対応する特徴語群を、加盟者端末装置3の使用者が手動操作で設定できるようにしてもよい。
図10に示す状態は、加盟者端末装置3の使用者が、区分を適宜指定した上で「ダウンロード」ボタンを押下することによって、プロトタイプデータベースサーバー装置4から、チャットパターンのプロトタイプをダウンロードし、さらに加盟者応答の欄にテキストを打ち込んだ状態である。ここに示す例では、加盟者端末装置3の使用者が区分1の「新幹線」を選択した場合を示している。ここに示す例では、「行き方」というタイトルのチャットパターンのプロトタイプがプロトタイプデータベースサーバー装置4から取得されている。代表入力として「行き方をおしえてください」というテキストが設定されている。類似入力として「どのようにしていけますか」、「お店までどうやっていけますか」、および「交通機関での行き方を教えてね」というテキストが設定されている。これら代表入力および類似入力のテキストは、プロトタイプデータとして、プロトタイプデータベースサーバー装置4の運営者により予めプロトタイプデータベースサーバー装置4内に書き込まれている。加盟者端末装置3の使用者は、これら代表入力および類似入力のテキストをそのまま使用することもできるし、編集して変更することもできる。
図11は、加盟者端末装置3において、類義語辞書を編集・登録するための画面の例を示す概略図である。図示するように、この類義語辞書登録画面は、検索語を入力する欄と「検索」ボタンとを有している。類義語辞書登録画面は、加キーワードと類義語の組を入力するための欄と、「登録」ボタンおよび「削除」ボタンとを、複数組分、有している。図示する例では、3組分が表示されているが、その数は任意に変更できるようにしてもよい。この画面の下方には、キーワードを追加するための欄(キーワード追加の欄と、その類義語の欄)が設けられている。さらに、この画面の下方には、キーワード追加に関して、「登録」ボタンと、「削除」ボタンと、「保存」ボタンとが設けられている。
図示する画面の状態の例では、検索語として「営業時間」が設定されている。
この加重値は、当該組における類義語を、キーワードの類義語として用いる度合いを表す。加重値の値が高いほど、キーワードの代わりに加重値が用いられる度合いが高いものとして扱われる。逆に、加重値の値が低いほど、キーワードの代わりに加重値が用いられる度合いが低いものとして扱われる。
同じく「削除」ボタンは、そのキーワードに対応する類義語を、類義語辞書から削除するためのトリガーとなるボタンである。
類義語辞書を用いた置換を許容することにより、入力されるテキストにおける表現の揺れを吸収して、テキストを解釈することが可能となる。類義語による置換を行う場合、前述の加重値を適切に設定することによって置換されやすさを制御することも可能である。
図12は、一般利用者端末装置7と自動応答サーバー装置2における仮想利用者との間で行われるチャットの実施の実例を示す概略図である。
同図(a)は、一般利用者端末装置7の外観(平面視)の一例を示す。同図(b)は、加盟者端末装置3のバーコード取得部36が取得したコード情報(アクセス情報)を印刷した物体(果実)の外観(斜視)の一例を示す。この印刷されるコード情報(アクセス情報)は、バーコードや、2次元コードや、文字や、その他の形態の情報であってよい。印刷されたコード情報は、例えば光学的に読み取り可能である。同図(c)は、一般利用者端末装置7側において表示されるチャットルームの画面例である。
(1)まず、不図示の加盟者端末装置3が、設定されたチャットパターンに基づいて学習処理を行う。そして、加盟者端末装置3は、自動応答サーバー装置2に対して、応答用知識データおよび入力解釈用知識データを送信する。
(2)自動応答サーバー装置2は、加盟者端末装置3側から送信された応答用知識データおよび入力解釈用知識データを記憶する。
(3)自動応答サーバー装置2は、当該加盟者のチャットサービスにアクセスするためのアクセス情報、または当該加盟者の特定の分野識別子に関連付けられたチャットサービスにアクセスするためのアクセス情報を、加盟者端末装置3に送信する。当該加盟者において、特定の分野識別子に関連付けられたチャットサービスにアクセスするようにするか否かは、例えば適宜設定可能とする。
(4)加盟者端末装置3は、自動応答サーバー装置2から受信したアクセス情報に基づいて、光学的に読み取り可能な2次元コードを印刷出力する。図12(b)において示される2次元コードがその例である。2次元コードが印刷された物体は、例えば、店舗等に置かれる。
(5)一般利用者端末装置7の使用者は、図12(a)に示す一般利用者端末装置7(具体的には、例えば、スマートフォンや、タブレット端末機など)を、図12(b)の2次元コードにかざすことにより、この2次元コードを読み取る。
(6)一般利用者端末装置7は、読み取った2次元コードから、チャットサービスにアクセスするための情報を抽出する。具体的には、一般利用者端末装置7は、例えば特定のチャットルームのサービスにアクセスするためのURL情報を抽出する。そして、一般利用者端末装置7は、通信手段を用いて、そのURLによって示されるサーバー装置にアクセスする。具体的には、そのURLは、チャットサーバー装置5における特定のサービスの所在を表す情報である。このようにして、一般利用者端末装置7は、加盟店に対応した、あるいは加盟店内の特定の情報に対応したチャットルームにアクセスすることができる。
この画面において、右側(利用者「戸田奈央子」と記載されている)は一般利用者端末装置7の利用者であり、左側(利用者「お店店主」と記載されている)は自動応答サーバー装置2における仮想利用者である。図示するチャットでは、まず、右側の一般利用者端末装置7の利用者が「子供のメニュもありますか?」と入力している。自動応答サーバー装置2の推論エンジン部21は、この入力テキストに基づき、最適な応答断片を推定する。
この推定の際には、予め学習処理によって得られたパラメーター値が使用される。パラメーター値は、自動応答サーバー装置2の入力解釈用知識データ記憶部22に記憶されている。そして、最適な応答断片が特定されると、自動応答サーバー装置2の推論エンジン部21は、その応答断片にしたがって応答を返す。具体的には、この例では、「はい!もちろんでございます。お子様用メニュも充実しております。是非、お子様とご一緒にどうぞ。」という応答のテキストを、自動応答サーバー装置2が返す。この応答のテキストは、チャットサーバー装置5を経由して一般利用者端末装置7に届けられ、その画面に表示される。その下に続く、一般利用者端末装置7からの入力である「そうですか?助かりました。ありがとう。」というテキストに対しても、同様の処理により、推論エンジン部は、最適な応答断片を推定し、その結果として「いいえ、どういたしまして。お気をつけてお越しください。」という応答のテキストを出力する。
まずステップ1として、加盟者が、自動応答サーバー装置2の運営者(以下、A社と言う)のホームページから会員登録をする。その加盟者が加盟会員と認められるとA社のプロトタイプデータベースサーバー装置4にアクセス可能となる。加盟者は、加盟者端末装置3を用いて、プロトタイプデータベースサーバー装置4から各産業分野別のツリー(分野ツリー)から、自らが該当するプロトタイプをダウンロードする。
(1)入力解釈用知識データ:入力テキストと応答との関係を表す知識(パラメーター)を含むデータ。
(2)応答用知識データ:上記代表入力及び類似入力に対する応答断片(答弁)を加盟者が入力したもののデータ。
以上の処理のこの流れにより、一般利用者は自分の知りたい内容を質問し、人工知能による処理を通じて必要な回答を得ることができる。
加盟者端末装置3は、プロトタイプデータベースサーバー装置4から、チャットパターンの原型となるデータをダウンロードできるため、手軽に、自事業の業種に合った応答システムを立ち上げることが可能となる。
次に、本発明の第2実施形態について説明する。前実施形態において既に説明した事項については以下において説明を省略する場合がある。ここでは、本実施形態に特有の事項を中心に説明する。
図示するように、応答システム101は、自動応答サーバー装置102と、加盟者端末装置3と、プロトタイプデータベースサーバー装置4と、チャットサーバー装置5と、一般利用者端末装置7とを含んで構成される。本実施形態の特徴は、前実施形態における自動応答サーバー装置2に代えて、自動応答サーバー装置102を含む点である。
これらのうち、推論エンジン部21と、入力解釈用知識データ記憶部22と、応答用知識データ記憶部23と、入力解釈用知識データ取得部24と、応答用知識データ取得部25との機能は、第1実施形態におけるそれらと同様である。本実施形態における自動応答サーバー装置102の特徴は、追加学習処理部128を有する点である。
次に、本発明の第3実施形態について説明する。前実施形態までにおいて既に説明した事項については以下において説明を省略する場合がある。ここでは、本実施形態に特有の事項を中心に説明する。
図示するように、応答システム201は、自動応答サーバー装置202と、加盟者端末装置3と、プロトタイプデータベースサーバー装置4と、チャットサーバー装置5と、一般利用者端末装置7とを含んで構成される。本実施形態の特徴は、前実施形態における自動応答サーバー装置102に代えて、自動応答サーバー装置202を含む点である。
上述した各実施形態では、チャットにおいて入力されるテキストの特徴として、そのテキストから抽出される特徴語を用いた。
類義語辞書を用いることによって、特徴語を拡張しても(置換しても)よいこととした。
変形例1では、入力テキストの特徴として、他の特徴を使用する。
例えば、入力テキストの構文解析あるいは係り受け解析を行い、解析結果であるツリー(構文解析ツリーあるいは係り受け解析ツリー)を特徴として使用する。このように、構文解析あるいは係り受け解析の結果を特徴として用いることにより、単純に特徴語を抽出する場合と比べて、入力テキストの内容を把握する精度が向上する場合がある。
例えば、第1の特徴語と第2の特徴語が入力テキスト内に出現する場合、それら2個の特徴語が、構文解析ツリーあるいは係り受け解析ツリー内の近い位置に共起するか、遠い位置で共起するかに依って、異なる特徴として扱うことができる。ここで、「近い」あるいは「遠い」ということは、例えば、ツリーのノード間の距離(ノードとノードとを結ぶアークのホップ数)により図られる。
上述した各実施形態では、自動応答サーバー装置2(または102)と、プロトタイプデータベースサーバー装置4と、チャットサーバー装置5とを、それぞれ独立の装置(コンピューター等)で構成した。
変形例2では、これらのサーバー装置のうちの複数を統合し、1台の装置内にそれら複数の機能を実装する。一例として、自動応答サーバー装置2(102)とプロトタイプデータベースサーバー装置4とを統合し、1台の装置とする。別の例として、自動応答サーバー装置2(102)とチャットサーバー装置5とを統合し、1台の装置とする。別の例として、プロトタイプデータベースサーバー装置4とチャットサーバー装置5とを統合し、1台の装置とする。自動応答サーバー装置2(102)とプロトタイプデータベースサーバー装置4とチャットサーバー装置5とを統合し、1台の装置としてもよい。
上述した実施形態では、類義語辞書を登録するようにしてもよかった。
変形例3では、代わりに、既存の類義語辞書データを予め用意しておき、推論エンジン部21がその類義語辞書データを参照できるようにする。この変形例3では、加盟者がいちいち個別に類義語辞書に該当する語の設定をする必要がなくなる。
上述した実施形態では、一般利用者端末装置7と推論エンジン部21における仮想ユーザーとが一対一でチャットを行う場合を説明した。変形例4では、三者以上の間でのチャットを行えるようにする。例えば、2台以上の一般利用者端末装置7(つまり、2人以上の一般利用者)と、推論エンジン部21における仮想ユーザーとがチャットを行う。この場合、チャットサーバー装置5は、三者以上のチャットをハンドリングする処理を行う。
本変形例の場合にも、推論エンジン部21が、知識に基づいて一般利用者端末装置7からの入力を解釈し、解釈結果に基づいて応答を返す処理を行うことに変わりはない。
Claims (13)
- 自動応答サーバー装置と、複数の端末装置とを具備する応答システムであって、
前記端末装置は、
チャットにおける前記自動応答サーバー装置への入力である入力断片と当該入力断片に対する前記自動応答サーバー装置からの応答である応答断片とのパターンであるチャットパターンの情報を生成するチャットパターン生成部と、
前記チャットパターン生成部で生成された前記チャットパターンに基づいて、前記入力断片に対応する入力テキストと前記応答断片との関係の学習処理を行い、前記学習処理の結果として得られる入力解釈用知識データを出力する学習処理部と、
前記チャットパターン生成部で生成された前記チャットパターンに含まれる前記応答断片に基づく応答用知識データを前記自動応答サーバー装置に送信する応答用知識データ送信部と、
前記学習処理部から出力された前記入力解釈用知識データを前記自動応答サーバー装置に送信する入力解釈用知識データ送信部と、
を具備し、
前記自動応答サーバー装置は、
前記端末装置の前記応答用知識データ送信部から送信された前記応答用知識データを記憶する応答用知識データ記憶部と、
前記端末装置の前記入力解釈用知識データ送信部から送信された前記入力解釈用知識データを記憶する入力解釈用知識データ記憶部と、
チャットにおける入力テキストと、前記入力解釈用知識データ記憶部に記憶された前記入力解釈用知識データとに基づき、前記応答用知識データ記憶部に記憶されている応答用知識データのうち、当該チャットにおける前記入力テキストに対応する前記応答断片を推定し、推定された前記応答断片に対応する応答用知識データを前記応答用知識データ記憶部から読み出すことによってチャットの応答を出力する推論エンジン部と、
を具備する、
ことを特徴とする応答システム。 - 前記チャットパターンの原型であるプロトタイプを記憶するプロトタイプ記憶部を備えたプロトタイプデータベースサーバー装置、
をさらに具備し、
前記チャットパターン生成部は、前記プロトタイプデータベースサーバー装置から取得した前記プロトタイプに基づいて、前記チャットパターンの情報を生成する、
ことを特徴とする請求項1に記載の応答システム。 - 前記プロトタイプデータベースサーバー装置は、
前記プロトタイプが属する分野の情報を管理する分野情報管理部、
を具備し、
前記プロトタイプは、前記分野と関連付けて管理される、
ことを特徴とする請求項2に記載の応答システム。 - 前記端末装置は、
前記学習処理部による前記学習処理が完了すると、当該学習処理の結果が反映されるチャットにアクセスするためのアクセス情報を取得するアクセス情報取得部、
をさらに具備する、
ことを特徴とする請求項1から3までのいずれか一項に記載の応答システム。 - 前記自動応答サーバー装置は、
光学的に読み取り可能な前記アクセス情報を含んだ画像情報を生成するアクセス情報生成部、
をさらに具備し、
前記アクセス情報取得部は、前記アクセス情報生成部によって生成された光学的に読み取り可能な前記アクセス情報を含んだ画像情報を取得する、
ことを特徴とする請求項4に記載の応答システム。 - 前記推論エンジン部は、利用者間のチャットのサービスを提供するチャットサーバー装置に接続することにより、前記チャットサーバー装置を経由して一般利用者端末装置との間でのチャットを実行するものであり、
チャットにおける前記入力テキストは、前記一般利用者端末装置から前記チャットサーバー装置を経由して前記推論エンジン部に送られるものであり、
チャットにおける前記応答は、前記推論エンジン部によって出力され、前記チャットサーバー装置を経由して前記一般利用者端末装置に送られるものである、
ことを特徴とする請求項1から5までのいずれか一項に記載の応答システム。 - 前記チャットサーバー装置を経由して前記推論エンジン部にチャットにおける前記入力テキストを送信するとともに、前記推論エンジン部から出力された前記応答を前記チャットサーバー装置を経由して受信する一般利用者端末装置、
をさらに具備することを特徴とする請求項6に記載の応答システム。 - 前記端末装置における前記アクセス情報取得部が取得した前記アクセス情報を、光学的な読取装置によって読み取るとともに、読み取った前記アクセス情報に基づいて前記チャットにアクセスし、チャットサーバー装置を経由して前記推論エンジン部にチャットにおける前記入力テキストを送信するとともに、前記推論エンジン部から出力された前記応答を前記チャットサーバー装置を経由して受信する一般利用者端末装置、
をさらに具備することを特徴とする請求項4または請求項5に記載の応答システム。 - 自動応答サーバー装置と、複数の端末装置とを用いた応答方法であって、
前記端末装置において、
チャットパターン生成部が、チャットにおける前記自動応答サーバー装置への入力である入力断片と当該入力断片に対する前記自動応答サーバー装置からの応答である応答断片とのパターンであるチャットパターンの情報を生成し、
学習処理部が、前記チャットパターン生成部で生成された前記チャットパターンに基づいて、前記入力断片に対応する入力テキストと前記応答断片との関係の学習処理を行い、前記学習処理の結果として得られる入力解釈用知識データを出力し、
応答用知識データ送信部が、前記チャットパターン生成部で生成された前記チャットパターンに含まれる前記応答断片に基づく応答用知識データを前記自動応答サーバー装置に送信し、
入力解釈用知識データ送信部が、前記学習処理部から出力された前記入力解釈用知識データを前記自動応答サーバー装置に送信し、
また、
前記自動応答サーバー装置において、
応答用知識データ記憶部に、前記端末装置の前記応答用知識データ送信部から送信された前記応答用知識データを記憶させ、
入力解釈用知識データ記憶部に、前記端末装置の前記入力解釈用知識データ送信部から送信された前記入力解釈用知識データを記憶させ、
推論エンジン部が、チャットにおける入力テキストと、前記入力解釈用知識データ記憶部に記憶された前記入力解釈用知識データとに基づき、前記応答用知識データ記憶部に記憶されている応答用知識データのうち、当該チャットにおける前記入力テキストに対応する前記応答断片を推定し、推定された前記応答断片に対応する応答用知識データを前記応答用知識データ記憶部から読み出すことによってチャットの応答を出力する、
ことを特徴とする応答方法。 - チャットにおける自動応答サーバー装置への入力である入力断片と当該入力断片に対する前記自動応答サーバー装置からの応答である応答断片とのパターンであるチャットパターンの情報を生成するチャットパターン生成部と、
前記チャットパターン生成部で生成された前記チャットパターンに基づいて、前記入力断片に対応する入力テキストと前記応答断片との関係の学習処理を行い、前記学習処理の結果として得られる入力解釈用知識データを出力する学習処理部と、
前記チャットパターン生成部で生成された前記チャットパターンに含まれる前記応答断片に基づく応答用知識データを前記自動応答サーバー装置に送信する応答用知識データ送信部と、
前記学習処理部から出力された前記入力解釈用知識データを前記自動応答サーバー装置に送信する入力解釈用知識データ送信部と、
を具備することを特徴とする端末装置。 - 外部の端末装置において生成されたチャットパターンに基づいて、チャットにおける入力である入力断片に対する応答である応答断片を、応答用知識データとして記憶する応答用知識データ記憶部と、
前記端末装置における学習処理によって生成された入力解釈用知識データであって、前記チャットパターンにおける前記入力断片に対応する入力テキストと、前記入力断片に対応する応答断片との関係に関して前記学習処理の結果として得られた入力解釈用知識データを記憶する入力解釈用知識データ記憶部と、
チャットにおける入力テキストと、前記入力解釈用知識データ記憶部に記憶された前記入力解釈用知識データとに基づき、前記応答用知識データ記憶部に記憶されている応答用知識データのうち、当該チャットにおける入力テキストに対応する前記応答断片を推定し、推定された前記応答断片に対応する応答用知識データを前記応答用知識データ記憶部から読み出すことによってチャットの応答を出力する推論エンジン部と、
を具備することを特徴とする自動応答サーバー装置。 - コンピューターを、
チャットにおける自動応答サーバー装置への入力である入力断片と当該入力断片に対する前記自動応答サーバー装置からの応答である応答断片とのパターンであるチャットパターンの情報を生成するチャットパターン生成部と、
前記チャットパターン生成部で生成された前記チャットパターンに基づいて、前記入力断片に対応する入力テキストと前記応答断片との関係の学習処理を行い、前記学習処理の結果として得られる入力解釈用知識データを出力する学習処理部と、
前記チャットパターン生成部で生成された前記チャットパターンに含まれる前記応答断片に基づく応答用知識データを前記自動応答サーバー装置に送信する応答用知識データ送信部と、
前記学習処理部から出力された前記入力解釈用知識データを前記自動応答サーバー装置に送信する入力解釈用知識データ送信部と、
を具備する端末装置として機能させるためのプログラム。 - 外部の端末装置において生成されたチャットパターンに基づいて、チャットにおける入力である入力断片に対する応答である応答断片を、応答用知識データとして記憶する応答用知識データ記憶部と、
前記端末装置における学習処理によって生成された入力解釈用知識データであって、前記チャットパターンにおける前記入力断片に対応する入力テキストと、前記入力断片に対応する応答断片との関係に関して前記学習処理の結果として得られた入力解釈用知識データを記憶する入力解釈用知識データ記憶部と、
を備えるコンピューターを、
チャットにおける入力テキストと、前記入力解釈用知識データ記憶部に記憶された前記入力解釈用知識データとに基づき、前記応答用知識データ記憶部に記憶されている応答用知識データのうち、当該チャットにおける入力テキストに対応する前記応答断片を推定し、推定された前記応答断片に対応する応答用知識データを前記応答用知識データ記憶部から読み出すことによってチャットの応答を出力する推論エンジン部、
を具備する自動応答サーバー装置として機能させるためのプログラム。
Priority Applications (9)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
MX2019015040A MX2019015040A (es) | 2017-07-14 | 2018-06-07 | Dispositivo de servidor de respuesta automatica, dispositivo de terminal, sistema de respuesta, metodo y programa de respuesta. |
EP18832858.7A EP3654211A4 (en) | 2017-07-14 | 2018-06-07 | AUTOMATED ANSWER DEVICE, TERMINAL DEVICE, ANSWER SYSTEM, ANSWER PROCEDURE AND PROGRAM |
US16/300,364 US10997371B2 (en) | 2017-07-14 | 2018-06-07 | Automatic response server device, terminal device, response system, response method, and program |
BR112019024578A BR112019024578A2 (pt) | 2017-07-14 | 2018-06-07 | dispositivo servidor de resposta automatizado, dispositivo terminal, sistema de resposta, método de resposta e programa |
CA3069568A CA3069568A1 (en) | 2017-07-14 | 2018-06-07 | Automated response server device, terminal device, response system, response method, and program |
RU2019136488A RU2745632C1 (ru) | 2017-07-14 | 2018-06-07 | Серверное устройство автоматизированного ответа, оконечное устройство, система ответа, способ ответа и программа |
SG11201912957UA SG11201912957UA (en) | 2017-07-14 | 2018-06-07 | Automated response server device, terminal device, response system, response method, and program |
CN201880028811.9A CN110582762A (zh) | 2017-07-14 | 2018-06-07 | 自动响应服务器装置、终端装置、响应系统、响应方法及程序 |
PH12019502894A PH12019502894A1 (en) | 2017-07-14 | 2019-12-20 | Automated response server device, terminal device, response system, response method, and program |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2017138423A JP6218057B1 (ja) | 2017-07-14 | 2017-07-14 | 自動応答サーバー装置、端末装置、応答システム、応答方法、およびプログラム |
JP2017-138423 | 2017-07-14 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2019012872A1 true WO2019012872A1 (ja) | 2019-01-17 |
Family
ID=60156816
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2018/021857 WO2019012872A1 (ja) | 2017-07-14 | 2018-06-07 | 自動応答サーバー装置、端末装置、応答システム、応答方法、およびプログラム |
Country Status (13)
Country | Link |
---|---|
US (1) | US10997371B2 (ja) |
EP (1) | EP3654211A4 (ja) |
JP (1) | JP6218057B1 (ja) |
KR (1) | KR101938790B1 (ja) |
CN (1) | CN110582762A (ja) |
BR (1) | BR112019024578A2 (ja) |
CA (1) | CA3069568A1 (ja) |
MX (1) | MX2019015040A (ja) |
PH (1) | PH12019502894A1 (ja) |
RU (1) | RU2745632C1 (ja) |
SG (1) | SG11201912957UA (ja) |
TW (1) | TW201908992A (ja) |
WO (1) | WO2019012872A1 (ja) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP6999987B1 (ja) * | 2021-04-12 | 2022-01-19 | 株式会社アクアビットスパイラルズ | アクション制御システム、アクション制御サーバ及びアクション制御方法 |
WO2022070542A1 (ja) | 2020-09-30 | 2022-04-07 | パナソニックIpマネジメント株式会社 | 電極、及び電極の製造方法 |
WO2022113498A1 (ja) | 2020-11-30 | 2022-06-02 | パナソニックIpマネジメント株式会社 | 電極、及び電極の製造方法 |
WO2022113682A1 (ja) | 2020-11-30 | 2022-06-02 | パナソニックIpマネジメント株式会社 | 電極、及び電極の製造方法 |
Families Citing this family (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9965783B2 (en) | 2014-02-07 | 2018-05-08 | Uber Technologies, Inc. | User controlled media for use with on-demand transport services |
AU2014386266A1 (en) | 2014-03-13 | 2016-09-29 | Uber Technologies, Inc. | Configurable push notifications for a transport service |
US9536271B2 (en) | 2014-05-16 | 2017-01-03 | Uber Technologies, Inc. | User-configurable indication device for use with an on-demand transport service |
US10212536B2 (en) | 2015-07-10 | 2019-02-19 | Uber Technologies, Inc. | Selecting a messaging protocol for transmitting data in connection with a location-based service |
US10325442B2 (en) | 2016-10-12 | 2019-06-18 | Uber Technologies, Inc. | Facilitating direct rider driver pairing for mass egress areas |
US10355788B2 (en) | 2017-01-06 | 2019-07-16 | Uber Technologies, Inc. | Method and system for ultrasonic proximity service |
JP6885506B2 (ja) * | 2018-03-02 | 2021-06-16 | 富士通株式会社 | 応答処理プログラム、応答処理方法、応答処理装置および応答処理システム |
US20190320043A1 (en) * | 2018-04-13 | 2019-10-17 | Uber Technologies, Inc. | Network computer system to generate synthetic messages based on service-specific information |
US11514915B2 (en) * | 2018-09-27 | 2022-11-29 | Salesforce.Com, Inc. | Global-to-local memory pointer networks for task-oriented dialogue |
US11398918B1 (en) | 2018-10-22 | 2022-07-26 | Wells Fargo Bank, N.A. | Passive user authentication |
JP6489670B1 (ja) * | 2018-10-30 | 2019-03-27 | Jeインターナショナル株式会社 | チャットシステム、チャットボットサーバー装置、チャットボットid管理装置、チャット仲介サーバー装置、プログラム、チャット方法、およびチャット仲介方法 |
JP7392382B2 (ja) * | 2018-11-30 | 2023-12-06 | 株式会社リコー | 情報処理システム、情報処理装置、情報処理方法及び情報処理プログラム |
JP7115265B2 (ja) | 2018-11-30 | 2022-08-09 | 富士通株式会社 | 対話制御方法、対話制御プログラム、対話制御装置、情報提示方法および情報提示装置 |
JP7230573B2 (ja) * | 2019-02-19 | 2023-03-01 | 株式会社リコー | 情報処理装置、情報処理システム、情報処理方法及びプログラム |
US20210004833A1 (en) | 2019-07-05 | 2021-01-07 | Talkdesk, Inc. | System and method for querying multiple information sources using agent assist within a cloud-based contact center |
US11328205B2 (en) | 2019-08-23 | 2022-05-10 | Talkdesk, Inc. | Generating featureless service provider matches |
US20210117882A1 (en) | 2019-10-16 | 2021-04-22 | Talkdesk, Inc | Systems and methods for workforce management system deployment |
US20210136220A1 (en) | 2019-10-31 | 2021-05-06 | Talkdesk, Inc. | Monitoring and listening tools across omni-channel inputs in a graphically interactive voice response system |
US11736615B2 (en) | 2020-01-16 | 2023-08-22 | Talkdesk, Inc. | Method, apparatus, and computer-readable medium for managing concurrent communications in a networked call center |
JP7466143B2 (ja) | 2020-03-16 | 2024-04-12 | カラクリ株式会社 | 情報処理装置、学習提案プログラム、及び学習提案方法 |
TWI771720B (zh) * | 2020-07-24 | 2022-07-21 | 華碩電腦股份有限公司 | 具有多型態輸入之辨識方法及使用其之電子裝置 |
TWI789626B (zh) * | 2020-09-04 | 2023-01-11 | 亞東學校財團法人亞東科技大學 | 情境引導化的適性語言學習系統及方法 |
US11677875B2 (en) | 2021-07-02 | 2023-06-13 | Talkdesk Inc. | Method and apparatus for automated quality management of communication records |
US11856140B2 (en) | 2022-03-07 | 2023-12-26 | Talkdesk, Inc. | Predictive communications system |
US11736616B1 (en) | 2022-05-27 | 2023-08-22 | Talkdesk, Inc. | Method and apparatus for automatically taking action based on the content of call center communications |
US11971908B2 (en) | 2022-06-17 | 2024-04-30 | Talkdesk, Inc. | Method and apparatus for detecting anomalies in communication data |
US11943391B1 (en) | 2022-12-13 | 2024-03-26 | Talkdesk, Inc. | Method and apparatus for routing communications within a contact center |
JP7338839B1 (ja) | 2023-02-12 | 2023-09-05 | 株式会社サマデイ | サーバ仲介システム、仲介サーバ、学習済モデルアシストシステム、及びアシストサーバ |
CN115952274B (zh) * | 2023-03-10 | 2023-06-27 | 北京百度网讯科技有限公司 | 基于深度学习模型的数据生成方法、训练方法和装置 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2004145541A (ja) * | 2002-10-23 | 2004-05-20 | Inosu:Kk | チャットシステム |
JP2016024765A (ja) * | 2014-07-24 | 2016-02-08 | ソフトバンク株式会社 | 情報検索装置及び情報検索プログラム |
JP2016045652A (ja) * | 2014-08-21 | 2016-04-04 | 国立研究開発法人情報通信研究機構 | 質問文生成装置及びコンピュータプログラム |
JP2016099967A (ja) * | 2014-11-26 | 2016-05-30 | 株式会社ユニバーサルエンターテインメント | 情報提供システム、情報提供方法、及びプログラム |
JP2017010517A (ja) | 2015-06-25 | 2017-01-12 | バイドゥ オンライン ネットワーク テクノロジー (ベイジン) カンパニー リミテッド | 人工知能によるヒューマン・マシン間の知能チャットの方法および装置 |
JP2017138423A (ja) | 2016-02-03 | 2017-08-10 | キヤノン株式会社 | 定着装置 |
Family Cites Families (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20030015696A (ko) * | 2001-08-17 | 2003-02-25 | (주)뉴타입테크놀러지 | 자동응답시스템과 연동하는 채팅 서비스 방법 |
JP2003108375A (ja) * | 2001-09-28 | 2003-04-11 | Seiko Epson Corp | 対話型エキスパートシステム及びプログラム |
US6946715B2 (en) * | 2003-02-19 | 2005-09-20 | Micron Technology, Inc. | CMOS image sensor and method of fabrication |
US7606714B2 (en) * | 2003-02-11 | 2009-10-20 | Microsoft Corporation | Natural language classification within an automated response system |
US20050054381A1 (en) * | 2003-09-05 | 2005-03-10 | Samsung Electronics Co., Ltd. | Proactive user interface |
JP5099367B2 (ja) * | 2006-03-10 | 2012-12-19 | 日本電気株式会社 | 要求内容識別システム、自然言語による要求内容の識別方法及びプログラム |
US20100240018A1 (en) * | 2007-01-30 | 2010-09-23 | Bethune Damion A | Process for creating and administrating tests |
JP2008191748A (ja) | 2007-02-01 | 2008-08-21 | Oki Electric Ind Co Ltd | ユーザ間コミュニケーション方法、ユーザ間コミュニケーションプログラム、ユーザ間コミュニケーション装置 |
CN100505664C (zh) * | 2007-03-27 | 2009-06-24 | 腾讯科技(深圳)有限公司 | 一种互动问答系统及其实现方法 |
KR101224660B1 (ko) * | 2008-07-09 | 2013-01-21 | 고려대학교 산학협력단 | 유사 문장 검색 장치, 검색 방법, 저장 매체, 자동 대화서비스 시스템 및 서비스 방법 |
US8818926B2 (en) * | 2009-09-29 | 2014-08-26 | Richard Scot Wallace | Method for personalizing chat bots |
WO2011094734A2 (en) * | 2010-02-01 | 2011-08-04 | Jumptap, Inc. | Integrated advertising system |
KR101521332B1 (ko) | 2011-11-08 | 2015-05-20 | 주식회사 다음카카오 | 인스턴트 메시징 서비스 및 인스턴트 메시징 서비스로부터 확장된 복수의 서비스들을 제공하는 방법 |
JP5825676B2 (ja) * | 2012-02-23 | 2015-12-02 | 国立研究開発法人情報通信研究機構 | ノン・ファクトイド型質問応答システム及びコンピュータプログラム |
US20170006161A9 (en) * | 2013-03-15 | 2017-01-05 | Genesys Telecommunications Laboratories, Inc. | Intelligent automated agent for a contact center |
KR101978793B1 (ko) | 2013-07-11 | 2019-05-15 | 삼성전자주식회사 | 차폐 장치 |
KR101590908B1 (ko) * | 2013-12-24 | 2016-02-03 | 서강대학교산학협력단 | 채팅 데이터 학습 및 서비스 방법 및 그에 따른 시스템 |
CN104572619A (zh) * | 2014-12-31 | 2015-04-29 | 天津市招财猫信息技术有限公司 | 智能机器人交互系统在投融资领域的应用 |
US10263929B2 (en) * | 2015-05-08 | 2019-04-16 | International Business Machines Corporation | Cloud based chat governance system based on behavioral patterns and situational-awareness |
US11516153B2 (en) * | 2016-01-25 | 2022-11-29 | Expensify, Inc. | Computer system providing a chat engine |
-
2017
- 2017-07-14 JP JP2017138423A patent/JP6218057B1/ja active Active
-
2018
- 2018-01-05 KR KR1020180001638A patent/KR101938790B1/ko active IP Right Grant
- 2018-06-07 WO PCT/JP2018/021857 patent/WO2019012872A1/ja unknown
- 2018-06-07 US US16/300,364 patent/US10997371B2/en active Active
- 2018-06-07 SG SG11201912957UA patent/SG11201912957UA/en unknown
- 2018-06-07 CA CA3069568A patent/CA3069568A1/en active Pending
- 2018-06-07 EP EP18832858.7A patent/EP3654211A4/en not_active Withdrawn
- 2018-06-07 RU RU2019136488A patent/RU2745632C1/ru active
- 2018-06-07 MX MX2019015040A patent/MX2019015040A/es unknown
- 2018-06-07 BR BR112019024578A patent/BR112019024578A2/pt unknown
- 2018-06-07 CN CN201880028811.9A patent/CN110582762A/zh active Pending
- 2018-07-10 TW TW107123857A patent/TW201908992A/zh unknown
-
2019
- 2019-12-20 PH PH12019502894A patent/PH12019502894A1/en unknown
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2004145541A (ja) * | 2002-10-23 | 2004-05-20 | Inosu:Kk | チャットシステム |
JP2016024765A (ja) * | 2014-07-24 | 2016-02-08 | ソフトバンク株式会社 | 情報検索装置及び情報検索プログラム |
JP2016045652A (ja) * | 2014-08-21 | 2016-04-04 | 国立研究開発法人情報通信研究機構 | 質問文生成装置及びコンピュータプログラム |
JP2016099967A (ja) * | 2014-11-26 | 2016-05-30 | 株式会社ユニバーサルエンターテインメント | 情報提供システム、情報提供方法、及びプログラム |
JP2017010517A (ja) | 2015-06-25 | 2017-01-12 | バイドゥ オンライン ネットワーク テクノロジー (ベイジン) カンパニー リミテッド | 人工知能によるヒューマン・マシン間の知能チャットの方法および装置 |
JP2017138423A (ja) | 2016-02-03 | 2017-08-10 | キヤノン株式会社 | 定着装置 |
Non-Patent Citations (1)
Title |
---|
See also references of EP3654211A4 |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2022070542A1 (ja) | 2020-09-30 | 2022-04-07 | パナソニックIpマネジメント株式会社 | 電極、及び電極の製造方法 |
WO2022113498A1 (ja) | 2020-11-30 | 2022-06-02 | パナソニックIpマネジメント株式会社 | 電極、及び電極の製造方法 |
WO2022113682A1 (ja) | 2020-11-30 | 2022-06-02 | パナソニックIpマネジメント株式会社 | 電極、及び電極の製造方法 |
JP6999987B1 (ja) * | 2021-04-12 | 2022-01-19 | 株式会社アクアビットスパイラルズ | アクション制御システム、アクション制御サーバ及びアクション制御方法 |
Also Published As
Publication number | Publication date |
---|---|
KR101938790B1 (ko) | 2019-01-16 |
TW201908992A (zh) | 2019-03-01 |
JP2019020995A (ja) | 2019-02-07 |
SG11201912957UA (en) | 2020-01-30 |
PH12019502894A1 (en) | 2020-12-07 |
CA3069568A1 (en) | 2019-01-17 |
CN110582762A (zh) | 2019-12-17 |
JP6218057B1 (ja) | 2017-10-25 |
MX2019015040A (es) | 2020-08-06 |
BR112019024578A2 (pt) | 2020-06-09 |
US20190147045A1 (en) | 2019-05-16 |
EP3654211A4 (en) | 2021-04-14 |
US10997371B2 (en) | 2021-05-04 |
EP3654211A1 (en) | 2020-05-20 |
RU2745632C1 (ru) | 2021-03-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6218057B1 (ja) | 自動応答サーバー装置、端末装置、応答システム、応答方法、およびプログラム | |
US11748555B2 (en) | Systems and methods for machine content generation | |
US11847106B2 (en) | Multi-service business platform system having entity resolution systems and methods | |
Wu et al. | Ai-generated content (aigc): A survey | |
US11823061B2 (en) | Systems and methods for continual updating of response generation by an artificial intelligence chatbot | |
US10664540B2 (en) | Domain specific natural language understanding of customer intent in self-help | |
US20220092028A1 (en) | Multi-service business platform system having custom object systems and methods | |
CN105760417A (zh) | 基于个性化用户模型和情境的认知交互式搜索的方法和系统 | |
CN109992763A (zh) | 语言标注处理方法、系统、电子设备及计算机可读介质 | |
JP2020071869A (ja) | 動画基盤求人求職マッチングサーバーおよび方法 | |
CN107590153A (zh) | 使用卷积神经网络的对话相关性建模 | |
US20210288927A1 (en) | Enabling communication with uniquely identifiable objects | |
US20230252224A1 (en) | Systems and methods for machine content generation | |
Paul et al. | Focused domain contextual AI chatbot framework for resource poor languages | |
Kumar et al. | Chatbot in Python | |
Alazzam et al. | Artificial intelligence chatbots: a survey of classical versus deep machine learning techniques | |
Khan et al. | Impact of Conversational and Generative AI Systems on Libraries: A Use Case Large Language Model (LLM) | |
CN108055192A (zh) | 群组生成方法、装置及系统 | |
US20230069587A1 (en) | Named entity recognition in chat dialogues for customer relationship management systems | |
Manaswi et al. | Developing chatbots | |
Bhardwaj et al. | Conversational AI—A State‐of‐the‐Art Review | |
CN109154899A (zh) | 用于利用在线服务执行用户指令的交互框架 | |
CN113011175A (zh) | 一种基于双重渠道特征匹配的语义识别方法及系统 | |
Zafar | Developing Smart Conversation Agent ECOM-BOT for Ecommerce Applications using Deep Learning and Pattern Matching | |
CN113836932A (zh) | 交互方法、装置和系统,以及智能设备 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 18832858 Country of ref document: EP Kind code of ref document: A1 |
|
REG | Reference to national code |
Ref country code: BR Ref legal event code: B01A Ref document number: 112019024578 Country of ref document: BR |
|
ENP | Entry into the national phase |
Ref document number: 3069568 Country of ref document: CA |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
ENP | Entry into the national phase |
Ref document number: 2018832858 Country of ref document: EP Effective date: 20200214 |
|
ENP | Entry into the national phase |
Ref document number: 112019024578 Country of ref document: BR Kind code of ref document: A2 Effective date: 20191122 |