Information processing method and device, electronic equipment and computer readable storage medium

Abstract

Embodiments of the present application provide an information processing method and device, electronic equipment and computer readable storage medium, relating to the technical field of natural language processing. The method comprises: obtaining attribute information corresponding to a robot according to current input of a user, then determining to-be-replied information corresponding to the current input of the user, and then generating a target reply based on the attribute information corresponding to the robot and the to-be-replied information corresponding to the current input of the user and outputting the target reply. Embodiments of the present application achieve more accurate output of a target reply when an information reply party (for example, an intelligent chat robot) interacts with a user, so as to realize a conversation with the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, in particular, the present application relates to an information processing method and device, electronic equipment and computer readable storage medium. BACKGROUND

[0002] With the development of information technology, intelligent chat robots have emerged, which can be widely used in multi-modal interaction. For example, an intelligent chat robot can directly dialogue with a user, or the intelligent chat robot can receive a question input by the user, retrieve an answer matching the question from a knowledge base, and output the answer to the user.

[0003] During information interaction between the intelligent chat robot and the user, when the intelligent chat robot receives input information of the user, the intelligent chat robot performs intent understanding on the input information of the user and generates a target reply to be output to the user. Therefore, when the intelligent chat robot is applied to interact with the user, how to accurately and individually output the reply to the user becomes a key problem. SUMMARY

[0004] The present application provides an information processing method, device, electronic equipment and computer readable storage medium, which is used to solve how to accurately output the reply information when the intelligent chat robot is applied to interact with the user. The technical solution is as follows:

[0005] In a first aspect, an information processing method is provided, which includes:

[0006] obtaining attribute information corresponding to the robot according to the current input of the user;

[0007] determining to-be-replied information corresponding to the current input of the user;

[0008] generating and outputting a target reply based on the attribute information corresponding to the robot and the to-be-replied information corresponding to the current input of the user.

[0009] In a second aspect, an information processing device is provided, which includes:

[0010] a first obtaining module, configured to obtain attribute information corresponding to the information reply party according to the current input of the user;

[0011] a determining module, configured to determine to-be-replied information corresponding to the current input of the user;

[0012] a first generating and outputting module, configured to generate and output a target reply based on the attribute information corresponding to the information reply party obtained by the first obtaining module and the to-be-replied information corresponding to the current input of the user determined by the determining module.

[0013] In a third aspect, an electronic device is provided, the electronic device comprising:

[0014] one or more processors;

[0015] a memory;

[0016] one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform operations corresponding to the information processing method according to the first aspect.

[0017] In a fourth aspect, a computer-readable storage medium is provided, the storage medium storing at least one instruction, at least one program, a code set, or an instruction set, the at least one instruction, the at least one program, the code set, or the instruction set being loaded and executed by a processor to implement the information processing method according to the first aspect.

[0018] In a fifth aspect, an information processing method is provided, the method comprising:

[0019] obtaining a current input of a user;

[0020] generating and outputting target reply information based on the current input of the user and historical interaction information.

[0021] In a sixth aspect, an information processing apparatus is provided, the apparatus comprising:

[0022] a second obtaining module configured to obtain a current input of a user;

[0023] a second generating and outputting module configured to generate and output target reply information based on the current input of the user and historical interaction information.

[0024] In a seventh aspect, an electronic device is provided, the electronic device comprising:

[0025] one or more processors;

[0026] a memory;

[0027] one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform operations corresponding to the information processing method according to the fifth aspect.

[0028] In an eighth aspect, a computer-readable storage medium is provided, the storage medium storing at least one instruction, at least one program, a code set, or an instruction set, the at least one instruction, the at least one program, the code set, or the instruction set being loaded and executed by a processor to implement the information processing method according to the fifth aspect.

[0029] The technical scheme provided by the present application has the beneficial effects that:

[0030] The present application provides an information processing method and device, electronic equipment and computer readable storage medium. Compared with the prior art, the present application determines the attribute information corresponding to the current information replier according to the current input of the user, then determines the information to be replied to the current input of the user, and then generates and outputs the target reply based on the attribute information corresponding to the current information replier and the information to be replied to the current input of the user. That is, the target reply output in the present application is generated based on the attribute information corresponding to the current information replier, and the attribute information corresponding to the current information replier is determined based on the current input of the user, so as to improve the accuracy of the target reply information output by the information replier (including intelligent chat robot), thereby improving the user experience.

[0031] The present application provides an information processing method and device, electronic equipment and computer readable storage medium. Compared with the prior art, the present application determines the attribute information corresponding to the current information replier according to the current input of the user, then determines the information to be replied to the current input of the user, and then generates and outputs the target reply based on the attribute information corresponding to the current information replier and the information to be replied to the current input of the user. That is, the target reply output in the present application is generated based on the attribute information corresponding to the current information replier, and the attribute information corresponding to the current information replier is determined based on the current input of the user, so as to improve the accuracy of the target reply information output by the information replier (including intelligent chat robot), thereby improving the user experience. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical scheme in the embodiments of the present application, the drawings needed in the description of the embodiments of the present application will be briefly introduced.

[0033] Figure 1a A chat scene conversation schematic diagram about personality information;

[0034] Figure 1b A chat scene conversation schematic diagram about emotional information;

[0035] Figure 1c A chat scene conversation schematic diagram about portrait information;

[0036] Figure 1d A chat scene conversation schematic diagram about historical interaction information;

[0037] Figure 1e Another chat scene conversation schematic diagram about portrait information;

[0038] Figure 2a A chat robot architecture schematic diagram provided by the embodiments of the present application.

[0039] Figure 2b A schematic diagram of a backend database provided for an embodiment of the present application;

[0040] Figure 2c A schematic diagram of attribute filling of an individual portrait database by a group database in an embodiment of the present application;

[0041] Figure 3 A schematic diagram of a method flow of information processing in an embodiment of the present application;

[0042] Figure 4 A schematic diagram of a method flow of information processing in another embodiment of the present application;

[0043] Figure 5a A schematic diagram of extracting historical knowledge from historical dialogues in an embodiment of the present application

[0044] Figure 5b An example diagram of storing historical knowledge information in an individual database in an embodiment of the present application;

[0045] Figure 6 A framework diagram of dialogue management and data updating based on hierarchical reinforcement learning in an embodiment of the present application;

[0046] Figure 7a A schematic diagram of a database updating strategy network in an embodiment of the present application;

[0047] Figure 7b A schematic diagram of a dialogue generation strategy network in an embodiment of the present application;

[0048] Figure 8 A framework diagram of NLG based on an encoding-decoding model in an embodiment of the present application;

[0049] Figure 9 A schematic diagram of a method based on tensor feature fusion in an embodiment of the present application;

[0050] Figure 10 A schematic diagram of a multi-task learning model framework in an embodiment of the present application;

[0051] Figure 11a A schematic diagram of a personality feature prediction model structure based on multi-label learning in an embodiment of the present application;

[0052] Figure 11b A schematic diagram of an emotional feature prediction model structure based on multi-label learning in an embodiment of the present application;

[0053] Figure 12 A schematic diagram of a flow of constructing a group database in an embodiment of the present application;

[0054] Figure 13Another positive growth chatbot architecture diagram provided by the embodiments of the present application;

[0055] Figure 14 A database update flow diagram in the embodiments of the present application;

[0056] Figure 15 An information processing device structure diagram in the embodiments of the present application;

[0057] Figure 16 Another information processing device structure diagram in the embodiments of the present application;

[0058] Figure 17 An electronic device structure diagram in the embodiments of the present application;

[0059] Figure 18 A block diagram of a computing system of an electronic device that can be used to implement the present application in the embodiments of the present application;

[0060] Figure 19a A user personality type and user personality template diagram in the embodiments of the present application;

[0061] Figure 19b A chatbot personality type and chatbot personality template diagram in the embodiments of the present application;

[0062] Figure 20a A user emotion type and user emotion template diagram in the embodiments of the present application;

[0063] Figure 20b A chatbot emotion type and chatbot emotion template diagram in the embodiments of the present application;

[0064] Figure 21a A user portrait attribute and user portrait template diagram in the embodiments of the present application;

[0065] Figure 21b A chatbot portrait attribute and chatbot portrait template diagram in the embodiments of the present application;

[0066] Figure 22a A user portrait information extraction flow diagram in the embodiments of the present application;

[0067] Figure 22b A chatbot portrait information extraction flow diagram in the embodiments of the present application;

[0068] Figure 23 A chatbot attribute prediction diagram in the embodiments of the present application. DETAILED DESCRIPTION

[0069] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting the invention.

[0070] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0071] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0072] With the development of artificial intelligence, intelligent chatbots have emerged. Chatbots can converse with users, including chatting and answering questions intelligently. More and more users now expect chatbots to have personalized conversations. While many chatbots exist, they have several shortcomings:

[0073] 1) In existing technologies, the personality and emotions of chatbots are predefined and remain unchanged during conversations;

[0074] For example, such as Figure 1a In the dialogue scenarios shown, two dialogue scenarios depict a mature adult and an immature adult talking to a chatbot when they are in a bad mood. However, the chatbot generates the same target response based on the same personality (predefined personality) and engages in dialogue with them. Obviously, this approach is not suitable for mature individuals, as they prefer to be listened to. As a result, the accuracy of the chatbot's dialogue with the user is low, leading to a poor user experience.

[0075] For example: Figure 1bIn the illustrated dialogue scenarios, when the user says "I will definitely win the championship tomorrow" to the Chatbot with serious and happy emotions (tones), the Chatbot replies "You will definitely fail" with a predefined emotion (a naughty tone). Obviously, when the user has a serious tone in the conversation with the Chatbot, the reply of the Chatbot is inappropriate and will make the user angry and lose interest in chatting.

[0076] In the above two dialogue scenarios, when different users have different personalities in the conversation with the Chatbot or the same user has different emotions in the conversation with the Chatbot, the Chatbot generates a target reply through a predefined personality or a predefined emotion to realize the conversation with the user, thereby resulting in low accuracy of the conversation between the Chatbot and the user, and further resulting in poor experience of the user in the process of the conversation with the Chatbot.

[0077] 2) In the prior art, only the user's portrait information is considered in personalizing the Chatbot, and the portrait information of the Chatbot itself is ignored.

[0078] For example, the user wants the Chatbot to have similar preferences as himself, such as Figure 1c In the two illustrated dialogue scenarios, a boy and a girl respectively have a dialogue with the Chatbot, and the question is "What is the recent movie you like?". For the boy, he generally prefers action movies, and the girl prefers romantic movies. However, the prior art does not consider the portrait information of the Chatbot itself, and the Chatbot may reply "I think XXX is very good, and the car racing is very exciting". Therefore, the accuracy of the conversation between the Chatbot and the user is low, and further, the user has poor experience in the process of the conversation with the Chatbot.

[0079] 3) The Chatbot does not have the ability of long-term memory, and cannot fully utilize the long-term historical information of the conversation in the dialogue with the user;

[0080] For example, as shown in Figure 1d When the user has a dialogue with the Chatbot on the first day, the user mentions a football match, and the next day, the user wants to continue the dialogue with the Chatbot about the football match yesterday, but the Chatbot has no background information of the dialogue yesterday, thereby resulting in that the Chatbot cannot give an accurate reply, and the user has poor experience;

[0081] 4) Task-oriented chatbots, such as MusicBots for the music industry, DoctorBots for the medical field, FoodBots for the food industry, and TelecomBots for e-commerce, have fixed and simple dialogue patterns, and generally have a limited number of sustainable dialogue rounds. Furthermore, they cannot engage in emotional or personalized conversations with users, resulting in poor flexibility.

[0082] 5) Existing technologies do not consider how to enable chatbots to continuously improve and grow through dialogue with users.

[0083] In order to solve the technical problems in the prior art, this application proposes the following solutions:

[0084] 1) Create a Chatbot that can have multiple different attribute information (personality, emotion, profile). When the Chatbot talks to the user, it first identifies the user's personality, emotion, user profile, etc., and selects the personality, emotion, and profile information of the chat partner that the user is most likely to like to have a conversation with the user.

[0085] 2) In the Chatbot system of this application embodiment, both user personalized features (attribute information) and Chatbot personalized features are considered, mainly including personality, emotion, profile, individual knowledge information, etc., and a joint feature learning method based on deep learning network is designed.

[0086] For example, taking user profile information as an example, a chatbot can consider user profiles when generating a target response, such as... Figure 1e The two dialogue scenarios shown represent male and female users interacting with the Chatbot, respectively, with the question "What are some good movies lately?". Male users generally prefer action movies, while female users tend to prefer romantic films. The response for male users is "Hmm, I think 'XXX' is excellent, the car chases are so exciting!" For female users, the Chatbot responds "Hmm, I think 'YYY' is very good, it's a very romantic and loving movie." This improves the accuracy of the Chatbot's responses and enhances the user experience.

[0087] 3) Introduce an individual knowledge base to extract key knowledge information from the dialogue text during the dialogue process and store it in the individual knowledge base to achieve long-term historical information memory; during the dialogue between the chatbot and the user, the chatbot can look up background information from the individual knowledge base to generate and output response information.

[0088] 4) Construct personalized and group databases to obtain robot attribute information related to user preferences through these databases:

[0089] The personalized database can include a user database and a chatbot database, both of which include a personality database, an emotion database, and a profile database.

[0090] The user database includes a personality database for storing user personality information, an emotion database for storing user emotion information, and a profile database for storing user profile information. The chatbot database includes a personality database for storing chatbot personality information, an emotion database for storing chatbot emotion information, and a profile database for storing chatbot profile information.

[0091] Both the user database and the chatbot database require defining a personality database template. This template consists of relevant personality categories, forming the coordinate basis of the personality database. The set of personality categories can be derived from behavioral psychology. Both the user database and the chatbot database use the same personality database template. The user database's personality database template is as follows: Figure 19a As shown, the personality database in the Chatbot database is as follows: Figure 19b As shown.

[0092] Both the user database and the chatbot database require defining an emotion database template. This template consists of relevant personality categories, forming the coordinate basis of the emotion database. The set of personality categories can be derived from behavioral psychology. Both the user database and the chatbot database use the same emotion database template. The emotion database template in the user database is as follows: Figure 20a As shown, the sentiment database in the Chatbot database is as follows: Figure 20b As shown.

[0093] Both the user database and the chatbot database require defining a profile database template. This template consists of relevant personality categories, forming the coordinate basis of the profile database. The set of personality categories can be derived from behavioral psychology. Both the user database and the chatbot database use the same profile database template. The user database's profile database template is as follows: Figure 21a As shown, the profile database in the Chatbot database is as follows: Figure 21b As shown.

[0094] 5) A personalized chatbot with positive growth is proposed. The personalization and positive growth of the chatbot are achieved by updating the personalized database and individual knowledge base in the dialogue. A database update method based on deep learning and hierarchical reinforcement learning is given.

[0095] The personalized database can include a user database and a chatbot database, both of which include a personality database, an emotion database, and a profile database.

[0096] Specifically, the user database includes a personality database for storing user personality information, an emotion database for storing user emotion information, and a profile database for storing user profile information; the chatbot database includes a personality database for storing chatbot personality information, an emotion database for storing chatbot emotion information, and a profile database for storing chatbot profile information.

[0097] In the embodiments of this application, the personality of the same user during the dialogue process can remain fixed, but the emotions of the same user during the dialogue process may change.

[0098] For example, personality can include: mature personality, sensitive personality, and stubborn personality; emotions can include: happy emotions, excited emotions, and mischievous emotions.

[0099] For the embodiments of this application, the profile information may include features such as social attributes, lifestyle habits, and personal preferences.

[0100] For example, the profile information includes at least one of the following: gender, age, weight, height, place of origin, occupation, educational background, and hobbies.

[0101] Based on the above solutions, this application provides an architecture for an actively evolving chatbot, used to implement a method for the chatbot to engage in dialogue with users and generate and output target responses:

[0102] Among them, such as Figure 2a As shown, this architecture mainly includes the following four basic modules, among which:

[0103] Natural Language Understanding (NLU) is primarily used for information extraction in dialogues and user intent recognition.

[0104] • Dialog Management (DM), the central component of the whole system, including Dialogue State Tracker (DST), Dialogue Policy and Database Policy;

[0105] 1) Dialogue State Tracker (DST), records the intermediate state of the system from the NLU module, such as the user intent category (Intent), the slot value filling (Slot Filling) confidence value distribution changes, the representation of the dialogue history background, the representation of the current utterance, etc., which are used for dialogue policy and database update policy generation; for example, "go to XXX for dinner", where the intent is "Food" and the slot value filling Adress = "XXX", where Adress is called a slot and "XXX" is a slot value. Slot value filling mainly refers to entity information extraction of input text. DST records Food (Adress = "XXX"), p = 0.8, where p refers to the confidence of this information, which can be understood as a prediction probability. This value will be continuously updated and confirmed as the conversation progresses.

[0106] 2) Dialogue Policy, which generates the probability distribution P(a|s) of all possible responses, where s is the current input state (including DST information and Chatbot personalized attribute information), and the state a is also called action, representing a possible generated response. Since the response can be a sentence of any length, the state space size of a is infinite;

[0107] 3) Database Policy (DB Policy), including four strategies: ① User Database Update Policy, which determines whether the user database needs to be updated; ② Chatbot Database Update Policy, which determines whether the Chatbot database needs to be updated; ③ Group Database Acquisition Policy, which determines whether to obtain Chatbot attribute information from the group database for updating the Chatbot database; ④ Individual Knowledge Base Update Policy, which decides whether to update the individual knowledge base, where individual knowledge is knowledge extraction from the entire dialogue input by NLU.

[0108] • Backend database and knowledge base, including individual database, group database and open knowledge base, Chatbot needs to interact with open database to obtain real-world knowledge.

[0109] • Natural Language Generation (NLG), generates responses based on dialogue background information, individual database and knowledge base.

[0110] The NLU obtains user utterance information of the user input, inputs the information understood in natural language to the DM module, the DM module interacts with the backend database and the knowledge base module to input the processing result to the natural language generation module NLG, the NLG generates reply information for the user utterance information of the user input based on the processing result of the DM and the information of the backend database and the knowledge base, and outputs to the user.

[0111] The backend database can include a group database, an individual database, and an open knowledge base, as shown in Figure 2b The group database can include a group personality database, a group emotion database, and a group portrait database, and the individual database includes a user database, a Chatbot database, and an individual knowledge base. The user database includes a user personality database, a user emotion database, and a user portrait database, and the Chatbot database includes a Chatbot personality database, a Chatbot emotion database, and a Chatbot portrait database. The user database is used to learn and update the individual characteristics of the user in the dialogue, and the Chatbot database is updated and adjusted according to the user individual information in the user database. The group database is mainly used to supplement and perfect the information of the user database and the Chatbot database, and the open knowledge base provides rich common sense knowledge for the Chatbot and can provide common sense knowledge for the individual knowledge base.

[0112] For the embodiments of the present application, the group database is a graph relationship database extracted based on a large amount of user dialogue data in the Internet, wherein the nodes of the graph relationship database represent users and their individual information (personality characteristics, emotional characteristics, and portrait characteristics, etc.), and the edges represent the preference relationship between users. Here, the preference relationship can be reflected according to the weight value obtained from the number of dialogues between users. The main functions of the group database are:

[0113] 1) Assist the user database to supplement the individual database information, that is, supplement and approximate the user individual information in the case that the user individual information obtained from the current dialogue information is incomplete (i.e. missing or the reliability is close to zero). Here, we take the group portrait database as an example to illustrate how the group database supplements the individual database information, as shown in Figure 2c As shown in the figure, in the user portrait database, user i lacks the attribute value of "interest", and in the group database, user k has the attribute value of "interest", and user k is similar to user i, so the attribute value of "interest" of user k can be used to fill in the missing information of user i.

[0114] 2) Auxiliary Chatbot individual database, on the basis of identifying user personalized features, help Chatbot select appropriate personalized features, i.e. the personalized features that the user may like the chat object to have. This operation is mainly to supplement from the group database in the case that the personalized features stored in the local Chatbot individual database are incomplete.

[0115] For the embodiments of the present application, the user personality database is used to store the corresponding personality features of the user, and can realize continuous learning and updating of the user personality features; the Chatbot personality database is used to store the corresponding personality features of the Chatbot, so that the Chatbot can have a dialogue with the user in different personalities; the user emotion database is used to store the corresponding emotion features of the user, and can realize continuous learning and updating of the user emotion features; the Chatbot emotion database is used to store the corresponding emotion features of the Chatbot, so that the Chatbot can have a dialogue with the user in different emotions; the user portrait database and the Chatbot portrait database are respectively used to store the portrait information of the user and the portrait information of the Chatbot. In the embodiments of the present application, the individual knowledge base is mainly used to record the local knowledge information in the dialogue, so as to realize the long-term background memory of the Chatbot. The database is connected with the dialogue management DM, and provides the dialogue background knowledge information for the NLG. The database is independent of other databases.

[0116] For the embodiments of the present application, the open database is also called online knowledge base or knowledge graph, which is mainly a kind of common sense knowledge network, such as <XXX, president, YY>, <AAA, height, 2.29m> respectively represent "XXX is the president of YY" and "the height of AAA is 2.29m". The individual database is connected with the open knowledge base, which can provide the Chatbot with rich common sense knowledge in the open domain.

[0117] In combination with the above structure, the embodiments of the present application provide an information processing method for realizing the dialogue between the user and the information reply party to solve the above technical problems. The method flow is as follows:

[0118] Step S101 (not shown in the figure), according to the current input of the user, the current corresponding attribute information of the information reply party is obtained.

[0119] For the embodiments of the present application, the information reply party can include: virtual robot and entity robot, wherein the virtual robot is an intelligent conversation system based on natural language processing, which is an intelligent chat robot integrating multiple artificial technologies; the entity robot is an entity device with virtual robot function.

[0120] Step S102 (not shown in the figure), determine the information to be replied to the user current input.

[0121] Specifically, the step S102 can include: determining the to-be-replied information corresponding to the current input of the user based on the attribute information currently corresponding to the information reply party.

[0122] The step S103 (not shown in the figure) generates a target reply based on the attribute information currently corresponding to the information reply party and the to-be-replied information corresponding to the current input of the user, and outputs the target reply.

[0123] The embodiments of the present application provide an information processing method. Compared with the prior art, the embodiments of the present application acquire attribute information currently corresponding to an information reply party according to current input of a user, then determine to-be-replied information corresponding to the current input of the user, and then generate a target reply based on the attribute information currently corresponding to the information reply party and the to-be-replied information corresponding to the current input of the user, and output the target reply. That is, the target reply output in the embodiments of the present application is generated based on the attribute information currently corresponding to the information reply party, and the attribute information currently corresponding to the information reply party is determined based on the current input of the user, so that the accuracy of the target reply information output by the information reply party (including an intelligent chat robot) can be improved, and the user experience can be improved.

[0124] In a possible implementation of the embodiments of the present application, the attribute information currently corresponding to the information reply party includes at least one of the following:

[0125] personality information; emotional information; portrait information;

[0126] The portrait information includes at least one of the following: gender, age, weight, height, place of origin, occupation, educational background, and hobbies.

[0127] In another possible implementation of the embodiments of the present application, the step S101 can specifically include: acquiring the attribute information currently corresponding to the information reply party according to historical interaction information and the current input of the user.

[0128] The historical interaction information includes historical interaction event information acquired from historical dialogue information.

[0129] In another possible implementation of the embodiments of the present application, acquiring the attribute information currently corresponding to the information reply party according to the current input of the user includes: if it is determined according to the current input of the user that the information reply party database is not to be updated, determining the attribute information currently corresponding to the information reply party as the attribute information of the information reply party corresponding to the last input of the user; and / or, if it is determined according to the current input of the user that the information reply party database is to be updated, updating the information reply party database according to the current input of the user, and acquiring the attribute information currently corresponding to the information reply party from the updated information reply party database.

[0130] In another possible implementation of the embodiments of the present application, the information replier database is updated according to the current input of the user, including: determining the predicted attribute information of the information replier according to the current input of the user, and updating the information replier database based on the predicted attribute information of the information replier.

[0131] In another possible implementation of the embodiments of the present application, the predicted attribute information of the information replier is determined according to the current input of the user, including: determining the attribute information currently corresponding to the user according to the current input of the user; and determining the predicted attribute information of the information replier according to the attribute information currently corresponding to the user.

[0132] In another possible implementation of the embodiments of the present application, the attribute information currently corresponding to the user is determined according to the current input of the user, including: determining the user attribute information corresponding to the current input; and determining the attribute information currently corresponding to the user based on the user attribute information corresponding to the current input and the user attribute information stored in the user database.

[0133] In another possible implementation of the embodiments of the present application, the predicted attribute information of the information replier is determined according to the attribute information currently corresponding to the user, including: determining the similar attribute information corresponding to the attribute information currently corresponding to the user; and determining the predicted attribute information of the information replier based on the determined similar attribute information corresponding to the attribute information currently corresponding to the user.

[0134] Specifically, the similar attribute information corresponding to the attribute information currently corresponding to the user is determined based on the attribute information currently corresponding to the user; the attribute information preferred by the user of the similar attribute information corresponding to the attribute information currently corresponding to the user is determined, and the attribute information preferred by the user of the similar attribute information corresponding to the attribute information currently corresponding to the user is determined as the predicted attribute information of the information replier.

[0135] In another possible implementation of the embodiments of the present application, the attribute information currently corresponding to the information replier is obtained from the updated information replier database, including: obtaining the attribute information of the information replier with the highest confidence from the updated information replier database as the attribute information currently corresponding to the information replier.

[0136] In another possible implementation of the embodiments of the present application, the attribute information currently corresponding to the user includes at least one of the following: personality information; emotional information; and portrait information.

[0137] The portrait information includes at least one of the following: gender, age, weight, height, place of birth, occupation, educational background, and hobbies.

[0138] In a possible implementation of the embodiment of the application, the personality information corresponding to the current user is determined based on the current input of the user, including: determining the probability distribution information of the user in each personality category based on the current input of the user; and determining the personality information corresponding to the current user based on the probability distribution information of the user in each personality category.

[0139] In a possible implementation of the embodiment of the application, the emotion information corresponding to the current user is determined based on the current input of the user, including: determining the probability distribution information of the user in each emotion category based on the current input of the user; and determining the emotion information corresponding to the current user based on the probability distribution information of the user in each emotion category.

[0140] The embodiment of the application provides another information processing method, which is used for implementing the conversation between the user and the information reply party to solve the above technical problem, and specifically as follows:

[0141] Step S201 (not shown in the figure), obtaining the current input of the user.

[0142] Step S202 (not shown in the figure), generating and outputting the target reply information based on the current input of the user and the historical interaction information.

[0143] The embodiment of the application provides an information processing method. Compared with the prior art, the embodiment of the application obtains the current input of the user, and then generates and outputs the target reply information based on the current input of the user and the historical knowledge information of the interaction between the user and the information reply party. That is, the target reply information output in the embodiment of the application is generated based on the current input of the user and the historical knowledge information of the interaction between the user and the information reply party, rather than only based on the current input of the user, so that the accuracy of the target reply information output by the information reply party (including an intelligent chat robot) can be improved, and the user experience can be improved.

[0144] In a possible implementation of the embodiment of the application, step S202 can include step S2021 (not shown in the figure) and step S2022 (not shown in the figure), wherein,

[0145] Step S2021, obtaining the historical knowledge information matched with the current input of the user.

[0146] Step S2022, generating and outputting the target reply information based on the current input of the user and the historical interaction information matched with the current input of the user.

[0147] The historical interaction information includes historical interaction event information obtained from the historical conversation information.

[0148] In another possible implementation of the embodiment of the application, the target reply information is generated and output, and then the user current input and the target reply information are extracted by a preset processing mode to obtain key information, and the key information is stored in a knowledge representation form.

[0149] The preset processing mode includes at least one of the following:

[0150] entity recognition, relationship extraction, event extraction, and noise reduction processing.

[0151] On the basis of the above embodiment, in combination with the above architecture, and through two specific examples (example one and example two), the specific process of generating the target reply by the chatbot (hereinafter referred to as the information reply party) and the user is introduced, example one: a personalized chat robot with positive growth is constructed, which can constantly improve and grow itself in the process of chatting with the user, and has personality, emotion, image, individual knowledge and other personalized characteristics, and can predict the attribute information (personalized characteristics) of the user in real time in the chatting process, and determine the attribute information of the Chatbot for chatting with the user according to the attribute information of the user, and chat with the user through the determined attribute information of the Chatbot; example two: the chat robot can have the ability of long-term background memory, store historical knowledge information (historical dialogue information of the user and the Chatbot in the interaction process), and generate a reply based on the stored historical knowledge information and output it to the user in the process of chatting with the user, as shown below:

[0152] Example one

[0153] The embodiment of the application introduces an information processing method, in which the attribute information of the Chatbot corresponding to the current input of the user (i.e., the personality, emotion and image of the Chatbot in the process of chatting with the user) can be determined, and the Chatbot is used to chat with the user in the process based on the attribute information of the Chatbot corresponding to the current input of the user. The specific method process is as shown in Figure 3

[0154] In step S301, the attribute information of the robot corresponding to the current input of the user is obtained.

[0155] For the embodiment of the application, the current input of the user includes: the current input of the user can include: text information of the current input of the user and / or voice information of the current input of the user.

[0156] In one possible implementation of the embodiment of the application, the attribute information of the robot corresponding to the current input of the user includes at least one of the following:

[0157] personality information, emotion information, image information,​

[0158] The portrait information includes at least one of the following: gender, age, weight, height, hometown, occupation, educational background, and hobby.

[0159] In another possible implementation of the embodiment of the application, the attribute information currently corresponding to the robot is acquired according to the current input of the user, including: if it is determined according to the current input of the user that the robot database is not to be updated, determining the attribute information currently corresponding to the robot as the attribute information of the robot corresponding to the last user input; and if it is determined according to the current input of the user that the robot database is to be updated, updating the robot database according to the current input of the user, and acquiring the attribute information currently corresponding to the robot from the updated robot database.

[0160] In another possible implementation of the embodiment of the application, the robot database is updated according to the current input of the user, including: determining the predicted attribute information of the robot according to the current input of the user, and updating the robot database based on the predicted attribute information of the robot.

[0161] In another possible implementation of the embodiment of the application, the predicted attribute information of the robot is determined according to the current input of the user, including: determining the attribute information currently corresponding to the user according to the current input of the user; and determining the predicted attribute information of the robot according to the attribute information currently corresponding to the user.

[0162] In another possible implementation of the embodiment of the application, the attribute information currently corresponding to the user includes at least one of the following:

[0163] personality information; emotional information; portrait information;

[0164] The portrait information includes at least one of the following: gender, age, weight, height, hometown, occupation, educational background, and hobby.

[0165] In another possible implementation of the embodiment of the application, the attribute information currently corresponding to the user is determined according to the current input of the user, including: determining the user attribute information corresponding to the current input; and determining the attribute information currently corresponding to the user based on the user attribute information corresponding to the current input and the user attribute information stored in the user database.

[0166] In another possible implementation of the embodiment of the application, the predicted attribute information of the robot is determined according to the attribute information currently corresponding to the user, including: searching for first similar attribute information based on the attribute information currently corresponding to the user, the first similar attribute information being attribute information similar to the attribute information currently corresponding to the user; searching for second similar attribute information, the second similar attribute information being the attribute information of the robot corresponding to the first similar attribute information; and determining the predicted attribute information of the robot as the second similar attribute information.

[0167] Another possible implementation of the embodiment of the application obtains the attribute information currently corresponding to the robot from the updated robot database, including: obtaining the updated robot attribute information from the updated robot database as the attribute information currently corresponding to the robot.

[0168] From the above embodiment, it can be known that in the process of obtaining the attribute information corresponding to the robot, the attribute information currently corresponding to the user needs to be determined based on the current input of the user, then it is determined whether the robot database is updated, and based on the determination result, the set of attribute information corresponding to the robot is obtained. The following describes how to determine the attribute information currently corresponding to the user based on the current input of the user, and then describes the process of determining whether the robot database is updated, and how to obtain the attribute information corresponding to the robot in the case of updating or not updating:

[0169] 1. Specifically, how to determine the attribute information (personality information, emotional information, and portrait information) currently corresponding to the user based on the current input of the user and the individual knowledge information (which can be collectively referred to as historical interaction events) such as entities, entity relationships, and intents is introduced:

[0170] For the embodiment of the application, a relatively direct way to determine the attribute information currently corresponding to the user based on the current input of the user is to establish a separate task learning model for each of these tasks (personality extraction task, emotion extraction task, portrait extraction task, and individual knowledge extraction task). This method is simple but ignores the fact that tasks are related and influence each other. Therefore, the embodiment of the application proposes a joint learning method (i.e., a deep neural network-based joint extraction model for extracting attribute information corresponding to the user), which regards these tasks as multi-task learning. Joint learning helps to complement information between tasks during feature extraction, which can improve the feature extraction effect of each task:

[0171] Specifically, public dialogue text is crawled from the network as dialogue training corpus, and a multi-task learning model constructed by training the dialogue training corpus. The model framework is as follows: Figure 10The network model mainly includes three layers: the first layer is an encoding layer, the input information includes background representation (user corresponding background representation and Chatbot corresponding background representation, user current input and the like; the second layer is a background information representation layer, that is, a context information feature extraction layer based on the input information; and the third layer is an output layer, outputting personality characteristics, emotional characteristics, portrait, individual knowledge, intent and the like. It should be noted that the joint extraction is mainly aimed at user input NLU analysis, and the individual knowledge in the output refers to entity, entity relationship, intent, slot value and the like, and the information in the individual knowledge base also includes event knowledge information, which is obtained from the user input and Chatbot reply in the historical dialogue, and the specific method is shown in Figure 5a and a method thereof are described.

[0172] For the embodiment of the present application, the personality characteristics extracted by the multi-task learning model mainly consist of some category labels and are represented in the form of a feature vector. For example, as shown in Figure 11a the text information corresponding to the current user input is "what interesting places can go?", and the multi-task learning model (which is an LSTM, CNN or DNN-based learning model) outputs the personality characteristic vector of the user (the probability of the user belonging to any personality type, including: the probability of the user belonging to the honest personality type is 0.6, the probability of the user belonging to the sensitive personality type is 0.7, and the probability of the user belonging to the prayer personality type is 0.9...).

[0173] For the embodiment of the present application, the selection of Chatbot personality characteristics is related to the personality of the user and the speech of the user, and a multi-label prediction model can be established to predict the personality of the Chatbot that the user may like, and the input information includes the personality characteristics of the current user, the current user speech and the Chatbot reply at the last time. Figure 23 The model for predicting the personality characteristics of the Chatbot is mainly divided into four layers:

[0174] The first layer is an input layer, including the current user speech (current user input) x1, the Chatbot reply at the last time x2 and the personality characteristics of the current user x3;

[0175] The second layer is to encode the three kinds of data information (current user input, Chatbot reply at the last time and current personality characteristics) by using neural networks (LSTM, CNN or NN) respectively to obtain h11, h12 and h13;

[0176] The third layer is a feature integration layer, which inputs the three kinds of encoded information h11, h12 and h13 into a neural network NN module (which can also include LSTM and CNN) for feature fusion;

[0177] The fourth layer is the output layer, which is a multi-label prediction output. Each node corresponds to a personality category, and the activation function is Sigmoid.

[0178] In the embodiments of this application, the sentiment features extracted by the multi-task learning model mainly consist of sentiment type labels and are represented in the form of feature vectors, for example, such as Figure 11b As shown, based on the user's current input speech, speech recognition is performed to obtain the corresponding text information "I will win the championship". Then, the emotional feature vector corresponding to the user is obtained through a multi-task learning model based on LSTM or CNN (the probability that the user belongs to any emotional type, including: the probability of belonging to the happy emotional type is 0.8, the probability of belonging to the mischievous emotional type is 0.9, the probability of belonging to the serious emotional type is 0.1, the probability of belonging to the crazy emotional type is 0.1, etc.).

[0179] In this embodiment, the selection of Chatbot emotional features is related to the user's emotions and speech. A multi-label prediction model can be established to predict the Chatbot emotional features the user might like. The input information includes the current user's emotional features, the current user's speech (current input), and the Chatbot's previous response, all jointly predicted. The network structure of the Chatbot emotional feature prediction model is similar to that of the Chatbot personality feature prediction network, and will not be elaborated here. It should be noted that no Chatbot speech is input when predicting Chatbot emotional information because the predicted user emotional features have already extracted the emotional information from the user's speech.

[0180] For the embodiments of this application, the user's current profile information can be based on a profile template such as... Figure 21a The data shown is extracted from relevant data. This relevant data may include user data, such as user text data in conversations, user app usage data, and external data. User profile information is extracted using methods such as feature extraction, clustering, classification, and dialogue analysis. Specifically, for example... Figure 22a As shown.

[0181] Furthermore, the profile information corresponding to the Chatbot can be based on the profile template (such as... Figure 21b (As shown) This data was extracted from relevant data. The data used to obtain the user profile includes user statements, chatbot response text data, operating environment data, and user feedback data. In addition, it also includes current user profile feature information. Feature extraction, clustering, and classification methods are used to extract the chatbot's profile information, specifically as follows... Figure 22b As shown.

[0182] 2. Specifically introduce whether the robot database is updated, and how to obtain the corresponding attribute information of the robot in the case of updating or not updating:

[0183] For the embodiments of the present application, the update strategy of the Chatbot database includes: a Chatbot database update strategy for determining whether to update the Chatbot database, and a group database acquisition strategy for determining whether to obtain Chatbot attribute information from the group database for updating the Chatbot database. In the embodiments of the present application, first, it is determined whether to update the Chatbot database according to the Chatbot database update strategy. If the Chatbot database update strategy indicates updating, the Chatbot database is updated. If the Chatbot database update strategy does not indicate updating, it is determined whether to update the Chatbot database according to the group database acquisition strategy, as shown in Figure 14 .

[0184] For the embodiments of the present application, if the Chatbot database update strategy and the group database acquisition strategy both indicate not updating the Chatbot database, it is determined that the attribute information currently corresponding to the robot corresponding to the current input of the user is the same as the attribute information of the robot corresponding to the last user input, and the target reply is still generated according to the attribute information of the robot corresponding to the last user input and output.

[0185] For the embodiments of the present application, according to the Chatbot database update strategy, the Chatbot database is updated, and then the predicted attribute information corresponding to the robot is determined based on the user's current input, and the robot database is updated through the determined predicted attribute information corresponding to the robot. If the Chatbot database update strategy does not indicate updating but the group database acquisition strategy indicates updating the robot database, the attribute information currently corresponding to the user is determined according to the user's current input, the attribute information similar to the attribute information currently corresponding to the user (first similar attribute information) is found from the group database according to the attribute information currently corresponding to the user, and then the attribute information corresponding to the robot corresponding to the first similar attribute information (second similar attribute information) is found from the group database based on the first similar attribute information, and then the robot database is updated through the second similar attribute information.

[0186] For the embodiments of the present application, if the Chatbot database is updated through the Chatbot database update strategy and / or the group database acquisition strategy, the set of updated Chatbot attribute information is input as the Chatbot attribute information on which the target reply is generated.

[0187] As described above, when updating the Chatbot database, it is necessary to supplement the user attribute information obtained based on the user's current input according to the user attribute database. Therefore, updating the user attribute database may be necessary during the process of obtaining Chatbot attribute information to generate the target response. In this embodiment, the user database update is determined based on the user database update strategy (determined by the database strategy in DM) and the completeness of the attribute information in the database.

[0188] Specifically, if the user database update policy instructs an update, the user database is updated based on the user attribute information obtained from the user's current input. If the user database update policy does not instruct an update, the decision to update the user database is based on the completeness of the attribute information in the user database (e.g., whether some attribute information is missing, or the probability of missing any attribute information). Figure 14 As shown, Figure 14 This is merely an example and is not intended to limit the update method. That is, the embodiments of this application may also update only the Chatbot database and not the user database, or update only the user database and not the Chatbot database.

[0189] In the embodiments of this application, when a user initiates an initial dialogue with a Chatbot, the user database and the Chatbot database may not store any attribute information. Therefore, when the user's current input is received, the robot attribute information corresponding to the user is determined through the group database.

[0190] Taking personality attributes as an example, this section explains how to determine the personality attributes of a chatbot corresponding to a user through a group database:

[0191] 1. Consider a user u with personality attribute X. When the action obtained from the group database retrieval strategy is yes, find the user u′ with the most similar personality to user u from the group personality database. When the action is no, do not update the Chatbot database and keep the Chatbot personality attribute unchanged, that is, still use the Chatbot personality attribute from the previous round of dialogue.

[0192] 2. Identify all personality traits {V1, V2, ..., V} that user u′ likes from the group personality database. n (This can be obtained based on the connectivity relationships in the population database graph);

[0193] 3. Calculate the similarity S(X, V) i Given i = 1, ..., n, select X′ = max which has the highest similarity to X.i S(X, V i ), as the personality attribute corresponding to the current chat Chatbot.

[0194] For the embodiments of the present application, the group database can include: a group personality database, a group emotion database, and a group portrait database. The embodiments of the present application can construct the group database in the following way, as shown in Figure 12

[0195] 1. Obtain a large number of open dialogue resources from the Internet;

[0196] 2. Use the portrait attribute extraction model, personality recognition model, and emotion recognition model to extract portrait, personality, and emotion feature information, respectively, and store them in the group portrait database, the group personality database, and the group emotion database;

[0197] 3. Extract the relationship between users, combine the above databases, and construct a portrait map, a personality map, and an emotion map, respectively;

[0198] 4. Cluster the group portrait (personality or emotion) and relationship preferences to obtain the characteristic information of the group, and construct the group database (group portrait database, group personality database, and group emotion database).

[0199] Step S302, determining the to-be-replied information corresponding to the current input of the user.

[0200] For the embodiments of the present application, step S302 can specifically include determining the to-be-replied information corresponding to the current input of the user based on the attribute information currently corresponding to the robot.

[0201] For the embodiments of the present application, step S301 introduces the database update strategy, and step S302 introduces the dialogue strategy, that is, determining the to-be-replied information corresponding to the current input based on the dialogue generation strategy. The dialogue generation is based on the database update, and there is a hierarchical relationship between the two. The embodiments of the present application realize the joint management of the two through hierarchical strategy reinforcement learning. The embodiments of the present application design two strategies for the hierarchical structure of generating to-be-replied information and database update, as shown in Figure 6 For each time step t = 1, …, T, assume that the dialogue history state sequence is s t = p1, q1, r1, …, p t , q t , r t , where p t , q t , r t represent the user utterance at time t, the Chatbot reply at time t, and the reward received by the Agent at time t, respectively.​t Two parts: reward r from database (individual database and individual knowledge base) update t1 and reward r from dialogue generation (generate target reply) t2 .

[0202] Specifically, the policy network of hierarchical policy reinforcement learning is composed of two sub-policy networks, i.e., database update policy network and dialogue generation policy network. In the reinforcement learning process, the two policy networks are coupled and jointly learned. In the embodiments of the present application, the database update policy network is as shown in Figure 7a , the goal is database update, and the database that needs to be updated is selected; the input is state s t1 , personality, emotion, image and historical features of local knowledge; the output is database strategy π(a) = P(a|s t1 ) for four databases; the network is deep neural network (DNN); the reward is the weighted sum of r t1 and r t2 , i.e., R t1 = λ1r t1 + λ2r t2 . In the embodiments of the present application, the dialogue policy network is constructed to generate dialogue generation strategy π(a) = P(a|s t2 ) for NLG dialogue generation, and the network structure is as shown in Figure 7b , wherein the goal is to select the next dialogue action and decide whether to select template generation or model generation for NLG; the input is state s t2 , including current utterance p t , background representation c t and Chatbot data d t ; the output is dialogue strategy π(a) = P(a|s t2 ) for all available reply actions; the network is an encoding and decoding network based on deep learning network (Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN) or Attention, etc.); the reward is reward r for evaluating the effect of dialogue generation. t2 .

[0203] For the embodiments of the present application, in reinforcement learning, reward is important information for updating the policy network. The definition of reward depends on the requirements of the actual task, so there can be many forms. The reward definition criteria of the database update policy network and the dialogue generation management policy network are introduced as follows:

[0204] • Database update reward r t1 Definition

[0205] (1) As low as possible database update frequency, such as

[0206] where represents the action k in state s t , is an indicator function.

[0207] (2) As high as possible increase in database information, such as

[0208] The average mutual information of two steps of states Φ (t-1)u and Φ tu before and after database update.

[0209] • Dialog generation reward r t2 Definition

[0210] (1) Information flow, Chatbot can generate information-rich replies in the conversation to ensure sustainable conversation, such as, r 21 = -log cos(h t · h t+1 ), where h t , h t+1 are the encoding representations from the encoder that generates consecutive utterances;

[0211] (2) Semantic relevance, the generated reply should be able to meet the syntax, while being related to the user utterance, such as, where the first term represents the probability of generating the reply a, and the second term represents the probability of generating the previous sentence of the conversation utterance.

[0212] The reward r t2 can be defined as a weighted sum of r 21 and r 22 , i.e., r t2 = u1r 21 + u2r 22 .

[0213] Step S303, based on the attribute information currently corresponding to the robot and the to-be-replied information corresponding to the current input of the user, a target reply is generated and outputted.

[0214] For the embodiments of the present application, based on the attribute information currently corresponding to the robot and the to-be-replied information corresponding to the current input of the user, a target reply is generated and outputted through a preset model.

[0215] Specifically, as Figure 8As shown, the preset model can include: an encoding-decoding model, the input data includes: the current input of the user, a dialogue generation strategy (dialogue strategy distribution), a background representation, attribute information corresponding to the current robot, and intent information, wherein the dialogue generation strategy is from a dialogue management module, the intent is from an NLU module, and the Chatbot data is from a Chatbot individual database and / or a group database. The output layer of the model not only generates a dialogue reply (target reply), but also generates a new background representation, which will be used as the background representation input of the next encoding-decoding model.

[0216] Further, as shown in Figure 8 , the current user input is input into an LSTM model, a CNN model or a Self-Attention model to obtain corresponding text information, and the obtained corresponding text information, dialogue strategy, background representation, Chatbot data, slot value and intent are fused through a fusion model, and then the fused information is input into an LSTM model, a CNN model or a Self-Attention model to obtain a target reply and a background representation input of the next encoding-decoding model.

[0217] The fusion module is mainly used to integrate speech features, database features and dialogue content features, and can be fused through the following formula:

[0218]

[0219] wherein h x represents the current input of the user, h y represents the background feature, h z represents the attribute information corresponding to the Chatbot, h m represents h x , h y , and h z represents the three modalities fused tensor, and "1" represents a vector with all values of 1. Considering the influence of bias, the dimension of the tensor is expanded; in the embodiment of the present application, the fused features can be one-dimensional tensors or multi-dimensional tensors.

[0220] Specifically, as shown in Figure 9 , h x , h y and h z are fused to obtain an expanded tensor feature h mThe tensor fusion technology can realize the fusion of multi-modal information, and can effectively obtain multi-dimensional interaction information between multiple modalities, such as single modal information, interaction information between two modalities, interaction information between three modalities, and the like. In actual application, complexity is avoided, and only interaction information of not more than N=3 modalities is taken (when N=1, there is no interaction at this time). Finally, the multiple fused tensor information is input into the decoding network. The fusion order of the modal fusion information and the modal information is irrelevant.

[0221] For the embodiment of the application, the target reply can also be generated by a template according to the slot value extracted from the current input of the user by the NLU and the knowledge, as shown in Table 1. The dialogue strategy network simultaneously judges whether to use the template generation. If Yes, the template generation is used, otherwise the model generation is used.

[0222] Table 1

[0223]

[0224] In Table 1, $V and $W can represent the slot value that needs to be filled, and $W represents that this is information that needs to be filled. The filled value is called a slot value relative to the slot. For example, area = "XXX", where area represents a slot, and "XXX" is a slot value obtained according to the input. The entire process is called slot value filling.

[0225] In another possible implementation of the embodiment of the application, step S301 can specifically include: obtaining attribute information corresponding to the robot at present according to at least one of the last input of the user and historical knowledge information and the current input of the user.

[0226] The historical knowledge information is historical interaction information between the user and the robot, and includes objects in historical dialogue information and relationship information between the objects.

[0227] For the embodiment of the application, attribute information corresponding to the robot at present can be obtained according to the last input of the user and the current input of the user, and can also be obtained according to the historical knowledge information and the current input of the user, and can also be obtained according to the last input of the user, the historical knowledge information, and the current input of the user. In the embodiment of the application, it is not limited.

[0228] In another possible implementation of the embodiment of the application, the personality information corresponding to the current user is determined based on the current input of the user, including: determining the probability distribution information of the user in each personality category based on the current input of the user; and determining the personality information corresponding to the current user based on the probability distribution information of the user in each personality category.

[0229] In another possible implementation of the embodiment of the application, the emotion information corresponding to the current user is determined based on the current input of the user, including: determining the probability distribution information of the user on each emotion category based on the current input of the user; and determining the emotion information corresponding to the current user based on the probability distribution information of the user on each emotion category.

[0230] Example Two

[0231] The embodiment of the application introduces an information processing method. In the embodiment, the process of the dialogue with the user can be performed according to the current input of the user and based on the historical knowledge information in the process of the interaction between the user and the robot. The example two can be combined with the example one to be implemented, or can be implemented alone. When the example two is implemented alone, the specific method flow is as shown in Figure 4

[0232] In step S401, the current input of the user is obtained.

[0233] For the embodiment of the application, the current input of the user can include the information input by the user in the text format, and can also include the information input by the user in the voice format. In the embodiment of the application, it is not limited.

[0234] In step S402, the target reply information is generated and output based on the current input of the user and the historical interaction information.

[0235] For the embodiment of the application, step S402 can specifically include: obtaining the historical knowledge information matched with the current input of the user; and generating and outputting the target reply information based on the current input of the user and the historical knowledge information matched with the current input of the user.

[0236] For the embodiment of the application, the historical knowledge information (historical interaction information) of the interaction between the user and the robot includes the objects in the historical dialogue information and the relationship information between the objects.

[0237] In the embodiment of the application, the historical knowledge information of the interaction between the user and the robot can be stored in the individual knowledge base constructed in advance. In the embodiment of the application, the current input of the user is obtained, the intent and other information corresponding to the current input of the user are determined based on the current input of the user, and the historical knowledge information matched with the current input of the user is obtained from the individual knowledge base according to the intent and other information corresponding to the current input. For example, the current input of the user is “yesterday's ball game was very interesting”, it is determined that the intent corresponding to the current user is “ball game” related information, and the historical knowledge information related to “ball game” is obtained from the individual knowledge base. Of course, since each piece of historical knowledge information in the individual knowledge base has a time stamp corresponding thereto, the time information can be combined when the historical knowledge information corresponding to the current input is obtained.

[0238] ​In another possible implementation of the embodiment of the application, the target reply information is generated and output, and then the current user input and the target reply information are extracted through a preset processing manner to obtain key information, and the key information is stored in the knowledge representation form.

[0239] The preset processing manner includes at least one of the following:

[0240] entity recognition, relationship extraction, event extraction, and noise reduction processing.

[0241] There are many mature methods for entity recognition and relationship extraction, such as a sequence labeling method based on deep learning for entity recognition, such as a bidirectional recurrent neural network (Bi-LSTM); a relationship classification method for relationship extraction based on entity recognition; and a joint learning method for entity recognition and relationship extraction, such as a CoType method, a Table Representation method, and the like. Event extraction mainly extracts event information mentioned in the dialogue, and entity information related to the event, such as a basketball game, event information including "team", "match time", "score", "match address", and the like, which can be learned by a sequence labeling method. The extracted information often has some errors or redundant information, which needs to be further processed.

[0242] Specifically, as shown in Figure 5a intent information is extracted from historical dialogue (user input information and robot reply information), including intent category (intent) and slot filling (slot filling) information, and then combined with an open database, and processed through entity recognition, relationship extraction, event extraction, and noise reduction processing, to obtain historical knowledge information, such as "time: T1, relationship: (athlete A, team, football team B), source: user", "time: T2, relationship: (athlete A, score, 3 balls), source: Chatbot", and the like, and the historical knowledge information is stored in the individual knowledge base.

[0243] For the embodiment of the application, the individual knowledge base records key information in the dialogue between the user and the Chatbot through the knowledge representation form, for example, "time: T1, relationship: (athlete A, team, football team B), source: user" and "time: T2, relationship: (athlete A, score, 3 balls), source: Chatbot" are stored in the individual knowledge base through the knowledge representation form, as shown in Figure 5b

[0244] ​The embodiment of the application provides an information processing method. Compared with the prior art, the embodiment of the application obtains current input of a user, and then generates and outputs target reply information based on the current input of the user and historical knowledge information of the user interacting with the robot. That is, the target reply information output in the embodiment of the application is generated based on the current input of the user and the historical knowledge information of the user interacting with the robot, instead of being generated only according to the current input of the user, so that the accuracy of the target reply information output by the robot can be improved, and in turn, the user experience can be improved.

[0245] On the basis of the above embodiment, the way of the user having a dialogue with the Chatbot is summarized and introduced, and specifically as shown in the following table, the architecture mainly includes four modules, namely natural language understanding (NLU), dialogue management (DM), individual database and group database (including open knowledge base), natural language generation (NLG) and the like. Figure 13 The architecture can support cloud (Cloud) end deployment, and can also support device end (On-Device) deployment.

[0246] Natural language understanding (NLU) is mainly to analyze the statement text in the dialogue, and the functions include extracting the intention, personality characteristics, emotional characteristics, image and individual knowledge implied in the dialogue of the user, and predicting the attribute characteristics of the chat object that the user may like. The input information is the current input (user utterance (Utterance) p) of the user and the last round Chatbot reply (Response) q, and the historical background representation generated by the NLG. The user personalized feature extraction and individual knowledge extraction mainly depend on the user utterance and the historical background representation, but in addition to the above information, the attribute information of the Chatbot that the user may like also needs to use the reply text of the last round Chatbot.

[0247] Dialogue management (DM) mainly includes the following three parts:

[0248] 1) Dialogue state tracking (DST), records the intermediate state of the information obtained by the system from the NLU module, such as the confidence value distribution change of the user intention (Intent) and slot filling (Slot Filling), dialogue historical background representation, current utterance representation (representation vector corresponding to the current input of the user) and the like, which is used for dialogue strategy and database update strategy generation.

[0249] 2) Dialogue Policy: This is the probability distribution P(a|s) for generating all possible responses, where s is the current input state (including DST information and Chatbot personalization information), and a, also called action, represents one possible response. Since responses can be sentences of arbitrary length, the state space of a is infinite.

[0250] 3) Database Policy, which includes four strategies: ① User database update strategy, which determines whether the individual user database needs to be updated; ② Chatbot database update strategy, which determines whether the individual chatbot database needs to be updated; ③ Group database acquisition strategy, which determines whether it is necessary to acquire personalized chatbot information from the group database in order to update the individual chatbot database; ④ Individual knowledge base update strategy, which determines whether the individual knowledge base needs to be updated, where individual knowledge comes from the knowledge extraction of the entire dialogue text by NLU.

[0251] Natural Language Generation (NLG) is the dialogue response text generation module, which employs template-based generation and neural network generation methods. The input information for this module includes the probability distribution of dialogue responses obtained according to the dialogue strategy, the background representation of the dialogue text, and personalized feature information obtained from the Chatbot database and individual knowledge bases.

[0252] The above embodiments describe the method for users to interact with a chatbot from the perspective of the process flow. The following description is from the perspective of a virtual module, as detailed below:

[0253] This application provides an information processing device, such as... Figure 15 As shown, the device 1500 may include: a first acquisition module 1501, a determination module 1502, and a first generation and output module 1503, wherein,

[0254] The first acquisition module 1501 is used to acquire the corresponding attribute information of the information responder based on the user's current input.

[0255] The determination module 1502 is used to determine the reply information corresponding to the user's current input.

[0256] The first generation and output module 1503 is used to generate and output a target reply based on the attribute information corresponding to the current responder obtained by the first acquisition module 1501 and the reply information corresponding to the current input of the user determined by the determination module 1502.

[0257] In a possible implementation of the embodiment of the application, the attribute information currently corresponding to the information reply party includes at least one of the following:

[0258] personality information, emotional information, and portrait information.

[0259] The portrait information includes at least one of the following: gender, age, weight, height, place of birth, occupation, educational background, and hobby.

[0260] In another possible implementation of the embodiment of the application, the first obtaining module is specifically configured to obtain the attribute information currently corresponding to the information reply party according to the current input of the user and historical interaction information.

[0261] The historical interaction information includes historical interaction event information obtained from the historical dialogue information.

[0262] In another possible implementation of the embodiment of the application, the first obtaining module 1501 is specifically configured to, when it is determined according to the current input of the user that the information reply party database is not to be updated, determine the attribute information currently corresponding to the information reply party as the attribute information of the information reply party corresponding to the last user input; and / or,

[0263] The first obtaining module 1501 includes an updating unit and a first obtaining unit, wherein,

[0264] The updating unit is configured to, when it is determined according to the current input of the user that the information reply party database is to be updated, update the information reply party database according to the current input of the user.

[0265] The first obtaining unit is configured to obtain the attribute information currently corresponding to the information reply party from the updated information reply party database.

[0266] In another possible implementation of the embodiment of the application, the updating unit includes a determining subunit and an updating subunit, wherein,

[0267] The determining subunit is configured to determine the predicted attribute information of the information reply party according to the current input of the user.

[0268] The updating subunit is configured to update the information reply party database based on the predicted attribute information of the information reply party.

[0269] In another possible implementation of the embodiment of the application, the determining subunit is specifically configured to determine the attribute information currently corresponding to the user according to the current input of the user.

[0270] The determining subunit is specifically further configured to determine the predicted attribute information of the information reply party according to the attribute information currently corresponding to the user.

[0271] In a possible implementation of the application, the determining subunit is specifically configured to determine the attribute information corresponding to the current user based on the attribute information corresponding to the current input and the attribute information of the user stored in the user database.

[0272] The determining subunit is specifically configured to determine the attribute information corresponding to the current user based on the attribute information corresponding to the current input and the attribute information of the user stored in the user database.

[0273] In a possible implementation of the application, the determining subunit is specifically configured to determine the similar attribute information corresponding to the current user based on the attribute information corresponding to the current user.

[0274] The determining subunit is specifically configured to determine the predicted attribute information of the information reply party based on the determined similar attribute information corresponding to the current user.

[0275] In a possible implementation of the application, the first obtaining unit is specifically configured to obtain the updated attribute information of the information reply party from the updated information reply party database as the attribute information currently corresponding to the information reply party.

[0276] In a possible implementation of the application, the attribute information currently corresponding to the user includes at least one of the following:

[0277] personality information, emotional information, and portrait information.

[0278] The portrait information includes at least one of the following: gender, age, weight, height, place of birth, occupation, educational background, and hobby.

[0279] In a possible implementation of the application, when the determining subunit is configured to determine the personality information corresponding to the current user based on the current input of the user, the determining subunit is specifically configured to determine the probability distribution information of the user in each personality category based on the current input of the user, and determine the personality information corresponding to the current user based on the probability distribution information of the user in each personality category; and / or,

[0280] When the determining subunit is configured to determine the emotional information corresponding to the current user based on the current input of the user, the determining subunit is specifically configured to determine the probability distribution information of the user in each emotional category based on the current input of the user, and determine the emotional information corresponding to the current user based on the probability distribution information of the user in each emotional category.

[0281] The embodiment of the present application provides an information processing device, compared with the prior art, the embodiment of the present application obtains attribute information currently corresponding to an information reply party according to current input of a user, then determines to-be-replied information corresponding to the current input of the user, and then generates a target reply based on the attribute information currently corresponding to the information reply party and the to-be-replied information corresponding to the current input of the user and outputs the target reply. That is, the target reply output in the embodiment of the present application is generated based on the attribute information currently corresponding to the information reply party, and the attribute information currently corresponding to the information reply party is determined based on the current input of the user, so that the accuracy of the target reply information output by the information reply party (including an intelligent chat robot) can be improved, and then the user experience can be improved.

[0282] The embodiment of the present application is applicable to the method embodiment, and details are not repeated here.

[0283] The embodiment of the present application provides another information processing device, as shown in Figure 16 The device 1600 can include a second acquisition module 1601 and a second generation and output module 1602, wherein

[0284] The second acquisition module 1601 is configured to acquire current input of a user.

[0285] The second generation and output module 1602 is configured to generate and output target reply information based on the current input of the user and historical interaction information.

[0286] Another possible implementation of the embodiment of the present application, the second generation and output module 1602 includes a second acquisition unit and a generation and output unit, wherein

[0287] The second acquisition unit is configured to acquire historical interaction information matched with the current input of the user.

[0288] The generation and output unit is configured to generate and output target reply information based on the current input of the user and the historical interaction information matched with the current input information of the user.

[0289] The historical interaction information includes historical interaction event information acquired from historical dialogue information.

[0290] Another possible implementation of the embodiment of the present application, the device 1600 further includes an extraction module and a storage module, wherein

[0291] The extraction module is configured to extract key information from the current input of the user and the target reply information through a preset processing mode.

[0292] The storage module is configured to store the key information extracted by the extraction module.

[0293] The preset processing mode includes at least one of the following:

[0294] entity recognition, relation extraction, event extraction, and noise reduction processing.

[0295] Embodiments of the present application provide an information processing device. Compared with the prior art, the embodiments of the present application obtain current input of a user, and then generate and output target reply information based on the current input of the user and historical knowledge information of interaction between the user and an information reply party. That is, the target reply information output in the embodiments of the present application is generated based on the current input of the user and the historical knowledge information of the interaction between the user and the information reply party, instead of being generated only according to the current input of the user, so that the accuracy of the target reply information output by the information reply party (including an intelligent chat robot) can be improved, and in turn, the user experience can be improved.

[0296] The information processing device provided by the embodiments of the present application is applicable to the method embodiments described above, and will not be described here again.

[0297] The information processing device provided by the embodiments of the present application is introduced from the perspective of functional modularization above. Next, the electronic device provided by the embodiments of the present application will be introduced from the perspective of hardware materialization, and the computing system of the electronic device will be introduced at the same time.

[0298] Embodiments of the present application provide an electronic device, which is applicable to the above-mentioned method embodiments, as shown in Figure 17 The electronic device includes a processor 1701 and a memory 1702 configured to store machine-readable instructions. When the instructions are executed by the processor 1701, the processor 1701 performs the above-mentioned information processing method.

[0299] Figure 18 A block diagram of a computing system of an electronic device that can be used to implement the present application according to embodiments of the present application is schematically shown. As shown in Figure 18 The computing system 1800 includes a processor 1810, a computer-readable storage medium 1820, an output interface 1830, and an input interface 1840. The computing system 1800 can execute the methods described above with reference to Figure 3 or Figure 4 to implement device control instructions based on user input to control at least one target device to perform corresponding operations. Specifically, the processor 1810 can include, for example, a general-purpose microprocessor, an instruction set processor, and / or a related chipset, and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1810 can also include on-board memory for cache purposes. The processor 1810 can be a single processing unit or a plurality of processing units for executing different actions of the method flows described with reference to Figure 3 or Figure 4 .

[0300] The computer readable storage medium 1820 may, for example, be any medium that can contain, store, communicate, propagate or transport instructions. For example, the readable storage medium may, for example, include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, apparatus, device or propagation medium. Specific examples of readable storage medium include a magnetic storage apparatus such as a hard disk (HDD), an optical storage apparatus such as a compact disc (CD-ROM), a memory such as a random access memory (RAM) or a flash memory, and / or a wired / wireless communication link.

[0301] The computer readable storage medium 1820 can include a computer program 1821, which can include code / computer executable instructions that, when executed by the processor 1810, cause the processor 1810 to perform, for example, the method processes described above in connection with Figure 3 or Figure 4 any variations thereof. The computer program 1821 can be configured to have, for example, computer program code including computer program modules. For example, in an example embodiment, the code in the computer program 1821 can include one or more program modules, for example, including 1821A, module 1821B, …. It should be noted that the division and number of modules are not fixed, and a person skilled in the art can use appropriate program modules or combinations of program modules according to actual conditions, and when these program modules or combinations of program modules are executed by the processor 1810, the processor 1810 can perform, for example, the method processes described above in connection with Figure 3 or Figure 4 any variations thereof.

[0302] According to embodiments of the present application, the processor 1810 can use the output interface 1830 and the input interface 1840 to perform the method processes described above in connection with Figure 3 or Figure 4 any variations thereof.

[0303] The embodiments of the present application provide an electronic device. Compared with the prior art, the embodiments of the present application determine the attribute information currently corresponding to the information reply party according to the current input of the user, then determine the to-be-replied information corresponding to the current input of the user, and then generate a target reply based on the attribute information currently corresponding to the information reply party and the to-be-replied information corresponding to the current input of the user and output the target reply. That is, the target reply output in the embodiments of the present application is generated based on the attribute information currently corresponding to the information reply party, and the attribute information currently corresponding to the information reply party is determined based on the current input of the user, so that the accuracy of the target reply information output by the information reply party (including an intelligent chat robot) can be improved, and the user experience can be improved.

[0304] The electronic device provided in the embodiments of the present application, compared with the prior art, obtains the current input of the user, and then generates and outputs the target reply information based on the current input of the user and the historical knowledge information of the interaction between the user and the information reply party. That is, the target reply information output in the embodiments of the present application is generated based on the current input of the user and the historical knowledge information of the interaction between the user and the information reply party, rather than only generated according to the current input of the user, so that the accuracy of the target reply information output by the information reply party (including an intelligent chat robot) can be improved, and in turn the user experience can be improved.

[0305] The electronic device and the computing system of the electronic device provided in the embodiments of the present application are applicable to the method embodiments described above, and will not be described herein again.

[0306] For the embodiments of the present application, the explanations of the same or similar terms in various embodiments can be mutually referred to, and will not be described again in each embodiment.

[0307] It should be understood that, although each step in the flowchart of the accompanying drawings is displayed in sequence according to the indication of the arrow, these steps are not necessarily executed in sequence according to the indication of the arrow. Unless explicitly described herein, the execution of these steps is not strictly limited in sequence, and can be executed in other sequences. Moreover, at least part of the steps in the flowchart of the accompanying drawings can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

[0308] The above only describes some embodiments of the present application, and it should be pointed out that, for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be regarded as the protection scope of the present application.

Claims

1. An information processing method, characterized in that, include: Retrieve the attribute information corresponding to the responder based on the user's current input; Determine the response information corresponding to the user's current input; as well as Based on the attribute information of the information responder and the information to be replied to corresponding to the current input of the user, generate the target reply and output it; The attribute information includes at least one of personality information and emotional information; The personality information is represented by a first feature vector; The emotional information is represented by a second feature vector; The personality information is stored in a personality database; The construction of the personality database defines a personality database template; The personality database template includes multiple personality categories to form the coordinate base of the personality database; The various personality categories mentioned are generalized based on behavioral psychology. The personality information includes the probability distribution information of the respondent in each personality category; The emotional information includes the probability distribution information of the respondent in each emotional category.

2. The method according to claim 1, characterized in that, The attribute information also includes portrait information.

3. The method according to claim 1 or 2, characterized in that, Based on the user's current input, retrieve the corresponding attribute information of the responder, including: Based on the user's current input and historical interaction information, obtain the attribute information corresponding to the information responder. The historical interaction information includes: historical interaction event information obtained from historical dialogue information.

4. The method according to claim 1, characterized in that, Based on the user's current input, obtain the attribute information corresponding to the responder, including at least one of the following: If it is determined not to update the information responder database based on the user's current input, then the current attribute information of the information responder is determined to be the same as the information responder attribute information corresponding to the user's previous input. If the user's current input determines that the information responder database needs to be updated, then the information responder database is updated according to the user's current input, and the attribute information corresponding to the information responder is retrieved from the updated information responder database.

5. The method according to claim 4, characterized in that, Update the database of information responders based on the user's current input, including: The predicted attribute information corresponding to the information responder is determined based on the user's current input, and the information responder database is updated based on the predicted attribute information corresponding to the information responder.

6. The method according to claim 5, characterized in that, Based on the user's current input, the predicted attribute information corresponding to the information responder is determined, including: Determine the user's current attribute information based on the user's current input; The predicted attribute information corresponding to the information responder is determined based on the user's current attribute information.

7. The method according to claim 6, characterized in that, The attribute information corresponding to the user is determined based on the user's current input, including: Determine the user attribute information corresponding to the current input; Based on the user attribute information corresponding to the current input and the user attribute information stored in the user database, the attribute information corresponding to the current user is determined.

8. The method according to claim 6 or 7, characterized in that, Determining the predicted attribute information corresponding to the information responder based on the user's current attribute information includes: Determine similar attribute information based on the user's current attribute information; Based on the identified similarity attribute information, the predicted attribute information corresponding to the information responder is determined.

9. The method according to claim 4, characterized in that, Retrieve the current attribute information of the information responder from the updated information responder database, including: The updated information responder attribute information is retrieved from the updated information responder database and used as the current attribute information corresponding to the information responder.

10. The method according to claim 6, characterized in that, The user's current attribute information includes at least one of the following: Personality information; emotional information; profile information.

11. The method according to claim 10, characterized in that, Based on the user's current input, determine the user's personality information, including: Based on the user's current input, determine the probability distribution information of the user in each personality category; The personality information corresponding to the current user is determined based on the probability distribution information of the user in each personality category. And / or, based on the user's current input, determine the emotional information corresponding to the current user, including: Based on the user's current input, determine the probability distribution information of the user in each emotion category; The emotional information corresponding to the current user is determined based on the probability information of the user in each emotional category.

12. The method according to claim 4, characterized in that, The step of retrieving the current attribute information of the information responder from the updated information responder database includes: When the updated information responder attribute information from the updated information responder database is incomplete, the attribute information corresponding to the current information responder is supplemented based on the attribute information from the group database.

13. The method according to claim 7, characterized in that, The step of determining the user's current attribute information based on the user attribute information corresponding to the current input and the user attribute information stored in the user database includes: When the user attribute information stored in the user database is incomplete, the attribute information corresponding to the current user is supplemented based on the attribute information from the group database.

14. The method according to claim 2 or 10, characterized in that, The profile information includes at least one of the following: gender, age, weight, height, place of origin, occupation, educational background, and hobbies.

15. The method according to claim 3, characterized in that, Based on the user's current input and historical interaction information, obtain the attribute information of the person responding to the message, including: Retrieve historical interaction information that matches the user's current input; Based on the user's current input and the historical interaction information that matches the user's current input, obtain the attribute information corresponding to the current responder.

16. The method according to claim 15, characterized in that, Generate and output the target response, followed by: Extract key information from the user's current input and the target response using a preset processing method, and store the key information; The preset processing method includes at least one of the following: Entity recognition; relation extraction; event extraction; noise reduction.

17. An electronic device, characterized in that, It includes: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: perform the information processing method according to any one of claims 1 to 16.

18. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement the information processing method as described in any one of claims 1 to 16.

Images