WO2018224034A1 - 智能问答方法、服务器、终端及存储介质 - Google Patents

智能问答方法、服务器、终端及存储介质 Download PDF

Info

Publication number
WO2018224034A1
WO2018224034A1 PCT/CN2018/090422 CN2018090422W WO2018224034A1 WO 2018224034 A1 WO2018224034 A1 WO 2018224034A1 CN 2018090422 W CN2018090422 W CN 2018090422W WO 2018224034 A1 WO2018224034 A1 WO 2018224034A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
corpus
question
reply
category
Prior art date
Application number
PCT/CN2018/090422
Other languages
English (en)
French (fr)
Inventor
郭振声
刘丹
陈维锋
苏志华
左堃田
罗潍红
王军伟
李文
Original Assignee
腾讯科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2018224034A1 publication Critical patent/WO2018224034A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/242—Query formulation
    • G06F16/243—Natural language query formulation

Definitions

  • the embodiments of the present invention relate to the field of Internet technologies, and in particular, to an intelligent question and answer method, a server, a terminal, and a storage medium.
  • the intelligent question answering system is a new type of information service system. Based on the functions of knowledge processing and semantic recognition, it can analyze user intentions and answer questions quickly and accurately. Because the intelligent question answering system can replace the dialogue between the real person and the user, and has the characteristics of rich knowledge and fast response, it is loved by the majority of users.
  • the intelligent question answering system can collect a large amount of conversation corpus information generated by the network users during the chat process, and train the question and answer model according to the collected conversation corpus information.
  • the reply information matching the question information can be generated based on the question and answer model and provided to the user. For example, when the user inputs the question information "How is the weather today", based on the question and answer model, the user can be provided with a matching reply message "except for smog, which is quite good".
  • the problem information proposed by the user may include multiple categories
  • the question and answer model trained according to the dialog corpus information in the above question and answer method is applicable to the dialogue scenario, and can be
  • the question information of the conversation category provides matching reply information, but it is difficult to provide matching reply information for other categories of question information. Therefore, the application scope of the above question and answer method is limited and not intelligent enough.
  • the embodiment of the invention provides an intelligent question and answer method, a server, a terminal and a storage medium, which can solve the problems of the related art.
  • the technical solution is as follows:
  • an intelligent question and answer method comprising:
  • the classification model is used to divide problem information of a professional category and problem information of a conversation category, where the problem information of the professional category refers to the content of the problem information Knowledge of the professional category, the problem information of the conversation category means that the content of the problem information does not involve knowledge of the professional category;
  • an intelligent question and answer method comprising:
  • the server is configured to determine, according to the classification model, a category to which the problem information belongs, when the problem information belongs to the professional category And obtaining, from the professional corpus database, reply information matching the problem information, wherein the classification model is used to divide the problem information of the professional category and the problem information of the dialogue category, and the professional corpus database is used to store the professional category
  • the corpus information, the problem information of the professional category means that the content of the problem information relates to knowledge of the professional category
  • the problem information of the conversation category means that the content of the problem information does not involve knowledge of the professional category ;
  • the reply information is displayed in the question and answer interface when receiving the reply information returned by the server.
  • an intelligent question answering device comprising:
  • a receiving module configured to receive a question and answer request sent by the terminal, where the question and answer request carries the problem information
  • a classification module configured to determine, according to the classification model, a category to which the problem information belongs, the classification model is used to divide problem information of a professional category and problem information of a conversation category, and the problem information of the professional category refers to the problem information
  • the content of the conversation relates to the knowledge of the professional category, and the problem information of the conversation category means that the content of the problem information does not involve the knowledge of the professional category;
  • a first reply module configured to: when the problem information belongs to the professional category, obtain reply information matching the problem information from a professional corpus database, where the professional corpus database is used to store corpus information of the professional category ;
  • a sending module configured to send the reply information to the terminal, where the terminal is configured to display the reply information.
  • an intelligent question answering device comprising:
  • a display module for displaying a question and answer interface of the smart question and answer application
  • a determining module configured to determine problem information to be replied in the Q&A interface
  • a sending module configured to send a question and answer request to the server by using the smart question and answer application, where the question and answer request carries the problem information
  • the server is configured to determine, according to a classification model, a category to which the problem information belongs, when the problem information When belonging to the professional category, the reply information matching the problem information is obtained from the professional corpus database, and the classification model is used to divide the problem information of the professional category and the problem information of the dialogue category, and the professional corpus database is used for Storing the corpus information of the professional category, where the content information of the professional category refers to the knowledge of the professional category, and the problem information of the conversation category refers to the content of the problem information does not involve Knowledge of professional categories;
  • the display module is further configured to display the reply information in the Q&A interface when receiving the reply information returned by the server.
  • a server comprising: a processor and a memory, wherein the memory stores at least one instruction, the instruction being loaded and executed by the processor to implement the following operations:
  • the classification model is used to divide problem information of a professional category and problem information of a conversation category, where the problem information of the professional category refers to the content of the problem information Knowledge of the professional category, the problem information of the conversation category means that the content of the problem information does not involve knowledge of the professional category;
  • a terminal comprising: a processor and a memory, wherein the memory stores at least one instruction, the instruction being loaded and executed by the processor to implement the following operations:
  • the server is configured to determine, according to the classification model, a category to which the problem information belongs, when the problem information belongs to the professional category And obtaining, from the professional corpus database, reply information matching the problem information, wherein the classification model is used to divide the problem information of the professional category and the problem information of the dialogue category, and the professional corpus database is used to store the professional category
  • the corpus information, the problem information of the professional category means that the content of the problem information relates to knowledge of the professional category
  • the problem information of the conversation category means that the content of the problem information does not involve knowledge of the professional category ;
  • the reply information is displayed in the question and answer interface when receiving the reply information returned by the server.
  • a seventh aspect a computer readable storage medium having stored therein at least one instruction loaded by a processor and executed to implement the smart question and answer method as described in the first aspect The action taken.
  • a computer readable storage medium stores at least one instruction loaded by a processor and executed to implement the smart question and answer method as described in the second aspect The action taken.
  • the method, the server, the terminal, and the storage medium provided by the embodiments of the present invention obtain a classification model, and the classification model is used to divide the problem information of the professional category and the problem information of the conversation category.
  • the classification is based on the classification.
  • the model determines the category to which the problem information belongs.
  • the matching reply information is obtained from the professional corpus database, and is sent to the terminal, which provides a way to reply to the problem information of the professional category, expands the application scope, and is based on
  • the classification model determines the category to which the problem information belongs, can identify the user's intention to ask the question, and then responds according to the user's intention, improves the accuracy of the reply, and improves the intelligence.
  • the relationship between the problem information and the multiple corpus information is calculated from the aspects of semantics, part of speech and syntactic structure, and the corpus information with the greatest relevance to the problem information is determined, and the corpus of the corpus information with the highest degree of relevance is matched.
  • the reply information the information realizes the sorting and sorting of the problem corpus information, and can more accurately answer the user's question.
  • the corresponding reply information is retrieved according to the user state information or the user attribute information, which can satisfy the personalized needs of the user, give an answer that meets the user's expectation, realize differentiated services for different user groups, and improve the pertinence.
  • the matching reply information is obtained based on the classification model, and the manner of replying to the problem information of the conversation category is provided, and the application scope is more comprehensive.
  • FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present invention.
  • FIG. 2 is a schematic diagram of an implementation environment provided by an embodiment of the present invention.
  • FIG. 3 is a flowchart of a method for intelligent question answering according to an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of an identifier list according to an embodiment of the present invention.
  • FIG. 5 is a schematic diagram of a question and answer interface according to an embodiment of the present invention.
  • FIG. 6 is a schematic diagram of a syntax structure provided by an embodiment of the present invention.
  • FIG. 7 is a schematic diagram of reply information provided by an embodiment of the present invention.
  • FIG. 8 is a schematic diagram of reply information provided by an embodiment of the present invention.
  • FIG. 9 is a schematic diagram of reply information provided by an embodiment of the present invention.
  • FIG. 10 is a schematic diagram of reply information provided by an embodiment of the present invention.
  • FIG. 11 is a schematic diagram of reply information provided by an embodiment of the present invention.
  • FIG. 12 is a schematic structural diagram of a server according to an embodiment of the present invention.
  • FIG. 13 is a schematic structural diagram of an intelligent question answering apparatus according to an embodiment of the present invention.
  • FIG. 14 is a schematic structural diagram of an intelligent question answering apparatus according to an embodiment of the present invention.
  • FIG. 15 is a schematic structural diagram of a terminal according to an embodiment of the present disclosure.
  • FIG. 16 is a schematic structural diagram of a server according to an embodiment of the present invention.
  • Intelligent Q&A application An application for replying to a question posed by a user, which may be an application running in an Internet application or an application client independent of an Internet application.
  • the developer can register the public identity for the smart question answering application in the internet application, and identify the smart question answering application with the public identity.
  • the public identity can be called a public number, a service number, or an enterprise number.
  • the ordinary user in the internet application can pay attention to the public identity of the smart question answering application or become a friend relationship with the public identity.
  • the Internet application can use the smart question-and-answer application running in the Internet application as a medium to realize the function of answering questions for ordinary users.
  • Internet application server refers to the server associated with the Internet application.
  • the Internet application server provides an open platform for providing an interface to a third-party server based on the original Internet application, and enhancing the functionality of the original Internet application or extending the platform for using the Internet application resource through access by a third-party server.
  • the internet application can be a social application, a payment application, or other related application.
  • Smart question and answer server refers to the third-party server associated with the smart question-and-answer application.
  • the intelligent question and answer server can access the Internet application server and implement the function of answering questions based on the functions or resources provided by the Internet application server.
  • FIG. 1 is a schematic diagram of an implementation environment according to an embodiment of the present invention.
  • the implementation environment includes: a terminal 101 and a server 102.
  • the terminal 101 and the server 102 are connected through a network.
  • the terminal 101 can be a mobile phone, a computer, a tablet computer, etc.
  • the server 102 can be a server, or a server cluster composed of several servers, or a cloud computing service center.
  • the terminal 101 is configured to determine the problem information to be replied and send it to the server 102.
  • the server 102 obtains the reply information of the question information and returns it to the terminal 101, thereby implementing a function for answering the question for the user.
  • the smart question and answer application associated with the server 102 is run in the terminal 101, and the smart question and answer application can interact with the server 102.
  • the smart question and answer application may be an application client installed on the terminal 101 or an application running in an internet application.
  • the smart question answering application is an application client installed independently on the terminal 101.
  • the smart question answering application is a front end application associated with the server 102, and the server 102 may be referred to as a smart question answering server.
  • the smart question answering application is an application running in an internet application.
  • the server 102 may include an internet application server 1021 and a smart question answering server 1022.
  • the terminal 101 and the internet application server 1021 pass through a network.
  • the connection is made between the Internet application server 1021 and the smart question answering server 1022 via a network.
  • the internet application is a front-end application associated with the internet application server 1021
  • the smart question-answer application is a front-end application associated with the smart question answering server 1022.
  • the Internet application server may be a social application server, a payment application, or the like. Accordingly, the Internet application server may be a social application server or a payment application server.
  • the terminal 101 can install an Internet application, and log in to the Internet application based on the user identifier to interact with the Internet application server 1021.
  • the user identifier can be a user account, a phone number, or the like.
  • the Internet application server 1021 is configured to provide an interface to the smart question answering server 1022 (a third-party server). After the smart question answering server 1022 accesses the Internet application server 1021, the smart question answering application server 1022 can register the public identifier on the Internet application server 1021, thereby The public identity interacts with the internet application server 1021 and uses the resources of the internet application server 1021 to answer questions for the user.
  • FIG. 3 is a flowchart of an intelligent question and answer method according to an embodiment of the present invention.
  • the smart question and answer method is applied to the implementation environment shown in the foregoing embodiment.
  • the interaction entity is a terminal and a server. Referring to FIG. 3, the method includes:
  • the terminal displays a question and answer interface of the smart question and answer application.
  • the Q&A interface may be a page in the form of HTML5 (Hyper Text Markup Language 5) or other forms of pages.
  • the smart question and answer application can be an application client installed on the terminal.
  • the operation of starting the smart question and answer application is triggered on the terminal.
  • the terminal detects the start operation the smart question and answer application is started, and the Q&A interface of the smart question and answer application is displayed.
  • the terminal installs an Internet application
  • the smart question answering application can be run in an internet application
  • the terminal displays an interface of the internet application
  • the internet application can be started in the interface of the internet application
  • the question and answer interface of the smart question and answer application is displayed.
  • the developer can register a public identity for the smart question-and-answer application, which serves as an entry point for the smart question-and-answer application to identify the smart question-and-answer application and also serves as a medium for the smart question-and-answer application, connecting the user to the smart question-and-answer application.
  • the public identity can be the name of the smart question and answer application, the application number, and the like.
  • the terminal can view the public identifier by using the Internet application and the user identifier based on the login.
  • the public identifier is included in the identifier list associated with the user identifier.
  • the user can view the public identifier.
  • the selection operation of the public identifier is triggered, and when the terminal detects the selection operation of the public identifier in the identifier list, the Q&A interface is displayed through the smart question and answer application.
  • the terminal may specifically pay attention to the public identity by searching the public identity in the Internet application, or pay attention to the public identity by scanning the two-dimensional code of the public identity in the Internet application.
  • the identifier list may include one or more public identifiers that the terminal is interested in, and may also include one or more user identifiers in the user relationship chain of the terminal, such as a user's friends, contacts, and the like, and may also include a server.
  • the service identifier provided for the terminal such as a weather forecast identifier.
  • the identifier list includes a user's friend and a public identifier of interest, wherein the public identifier “micro-question” is used to answer questions for the user, and when the user clicks on the public identifier “micro-question”, the display is as shown in FIG. 5 .
  • the question and answer interface shown.
  • the terminal determines the problem information to be replied in the question and answer interface, and sends a question and answer request to the server through the smart question and answer application.
  • the question and answer request carries question information, and the question and answer request is used to request the server to reply to the question information.
  • the user can input the problem information in the question and answer interface, and the terminal obtains the input question information.
  • the problem information can include a variety of formats such as text, pictures, video, audio, and the like.
  • the input method is different.
  • the question and answer interface includes an input field, and the user can input the question information in the text format in the input field; the question and answer interface may further include an audio input button, and the user may input the question information of the audio format when pressing the audio input button.
  • the problem information may also include various categories, such as news, computer technology, chemistry, securities, dialogue, etc., and may include problem information of a professional category, and problem information of a conversation category, and specific categories of problem information. Determined based on the content of the problem information.
  • the terminal displays a list of problem information in the Q&A interface, where the problem information list includes a plurality of problem information that has been set, and the user can select the problem information to be answered, and the terminal selects the problem selected by the user. information.
  • the problem information list may be set by the server by default, or may be determined after statistics of the problem information with a large number of occurrences.
  • the terminal displays the problem information list and the input field in the question and answer interface, and the user may select the problem information from the problem information list, or may also answer questions. Enter the problem information in the interface.
  • the server receives the question and answer request, determining, according to the classification model, the category to which the problem information belongs.
  • the server is a smart question answering server associated with the smart question answering application.
  • the server when the smart question answering application runs in an internet application, the server is associated with an internet application and is also associated with a smart question answering application.
  • the server can include an internet application server associated with the internet application and a smart question answering server associated with the smart question answering application.
  • the server may reply to the problem information of the professional category and the problem information of the conversation category.
  • the problem information of the professional category refers to the content of the problem information belonging to the professional category, and needs to be answered according to the knowledge of the professional category, which may include one or more categories such as computer technology, securities, and chemistry.
  • the problem information of the conversation category means that the content of the problem information belongs to the daily conversation content, and does not involve the knowledge of the professional category.
  • the embodiment of the present invention provides different reply modes.
  • the problem information raised by the user is the problem information of the professional category
  • the user provides professional reply information according to the relevant knowledge of the professional category.
  • the question information raised by the user is the question information of the conversation category
  • a chat conversation with the user is made.
  • the server when the server receives the question and answer request, it first determines the category to which the problem information belongs based on the classification model, and then can reply according to the reply mode corresponding to the category to which the problem information belongs.
  • the classification model is used to divide the problem information of the professional category and the problem information of the conversation category, and according to the classification model, it can be determined whether the problem information belongs to a professional category or a conversation category.
  • the classification model can be obtained by using a machine learning algorithm according to the collected plurality of sample problem information and the category to which it belongs, and the machine learning algorithm can be a support vector machine algorithm or other algorithms.
  • multiple sample problem information and its associated categories may be collected, and each sample problem information is segmented to obtain multiple phrases of each sample problem information, according to The feature of the phrase constructs a feature vector of each sample question information, thereby training the classification model according to the feature vector of the plurality of sample question information and the associated category, and the classification model can determine the corresponding category according to the feature vector of the problem information.
  • the problem information may be segmented first, and multiple phrases in the question information are obtained, and the feature vector is constructed according to the features of each phrase in the plurality of phrases, and the feature vector is input into the classification model. Then, the category of the feature vector can be determined based on the classification model, which is the category to which the problem information belongs.
  • the above word segmentation process may be performed based on a word segmentation model, which may be a conditional random field model, a hidden Markov model, or the like.
  • a word segmentation model which may be a conditional random field model, a hidden Markov model, or the like.
  • the question information is "what is available funds”
  • multiple phrases can be obtained: “what”, “yes”, “available”, “funds”.
  • the server may set a synonym database in which the synonym database corresponding to some phrases is set in the synonym database, in consideration of the fact that the user may input the problem information and may include the wrong phrase in the problem information. It is the correct phrase obtained after the phrase is corrected. Then, before the problem information is segmented, the server may first replace the problem information with the synonym, and replace the phrase included in the question information with the corresponding synonym according to the synonym database, and then perform the word segmentation. For example, the question information input by the user is "how much the bank evaluates", and at this time, the problem information can be automatically corrected to "how much the bank stock price". Synonym replacement can identify the user's typos, implement error correction, and locate the problem information in the correct semantic context to give a response that meets the user's expectations.
  • the problem information contains some auxiliary words without actual semantics
  • the phrase with actual semantics can be retained, and other phrases can be removed.
  • the feature vector is constructed according to the features of the reserved phrase.
  • the embodiment of the present invention classifies the problem information by using the classification model, and can identify whether the user's intention is to consult the professional category question or to chat, so as to give the user a satisfactory answer according to the user's intention. Further, using a machine learning algorithm to train the classification model can improve accuracy and more accurately understand the user's intention.
  • the server obtains the reply information that matches the problem information from the professional corpus database, and performs step 306.
  • the retrieval mode is used to obtain the reply information, and the retrieval mode refers to setting the professional corpus database, and after receiving the input problem information, the reply information is extracted by searching and matching in the professional corpus database.
  • the professional corpus database is used to store the corpus information of the professional category, and the corpus information may include various types such as text information, picture information, video information, audio information, and link information.
  • the corpus information in the professional corpus database may be obtained from the Internet by using a crawler downloading technology, or may be set according to corpus information that may be involved in the function provided by the smart question answering application, or may be obtained by other means, and in the application process. It is also possible to continuously accumulate problems raised by users, expand the relevant knowledge of professional categories, and update the professional corpus database so that the corpus information is more and more comprehensive, and the follow-up can more accurately answer questions for users. After obtaining the corpus information, it can be manually reviewed to ensure the quality of the corpus information. For the wrong corpus information, you can edit and correct the error manually to get the correct corpus information.
  • the corpus information obtained by the crawler downloading technique can be as shown in Table 1 below.
  • the corpus information in the professional corpus database may include problem corpus information and reply corpus information that match each other, that is, the question template and the corresponding answer may be preset, and the corpus information is stored in the professional corpus database. Then, when the server determines that the problem information belongs to the professional category, the problem corpus information matching the problem information may be retrieved in the professional corpus database, so that the reply corpus information matching the problem corpus information is used as the reply information of the question information.
  • the problem corpus information and the reply corpus information in the corpus information can be as shown in Table 2 below.
  • the reply corpus information matching any question corpus information may be used as the reply information of the question information, or may be obtained from multiple question corpus information.
  • the degree of relevance of each problem corpus information to problem information indicates the degree of similarity between the problem corpus information and the problem information. The greater the degree of association, the more similar the problem corpus information is to the problem information, and the more likely it is the same problem. Therefore, determining the problem corpus information with the greatest degree of relevance to the problem information in the plurality of problem corpus information can be regarded as the problem corpus information most similar to the problem information, and determining the response corpus information matching the problem corpus information with the highest degree of relevance. A reply message that matches the problem information.
  • At least one of the semantic relevance degree, the part-of-speech relevance degree and the syntactic structure relevance degree of the problem corpus information and the problem information may be obtained, according to the semantic relevance degree, the part-of-speech relevance degree and At least one of the syntactic structure relevance degrees obtains the degree of association between the problem corpus information and the problem information.
  • the semantic relevance degree is used to indicate the semantic similarity between the problem corpus information and the problem information. The greater the semantic relevance, the more likely the semantics of the problem corpus information and the problem information are the same.
  • the semantic relevance degree is calculated according to the similarity between the phrase in the problem corpus information and the phrase in the question information.
  • the similarity between any two phrases may be the reciprocal of the distance between the overall feature vectors of the two phrases, and the overall feature vector of the phrase refers to a vector consisting of the features of the phrase in multiple dimensions, such as The vector consisting of part-of-speech features, semantic features, and appearance characteristics of the phrase, the smaller the overall feature vector distance of the two phrases indicates that the more similar the two overall feature vectors are, the more similar the two phrases are.
  • the similarity between “Shanghai Stock Market” and “Shanghai” is 0.85
  • the similarity between “Shanghai Stock Market” and “Shenzhen City” is 0.8, indicating that the terms “Shanghai Stock Market” and “Shanghai” are more similar.
  • the part of speech relevance is used to indicate the degree of similarity between the phrase in the problem corpus information and the phrase of the question information.
  • the greater the degree of relevance of the part of speech the more likely the part of the corpus information in the problem information and the question information is the same, then the problem corpus information and problem The more likely the information is the same problem.
  • the problem corpus information and the problem information can be segmented to obtain the part of speech of each phrase in the problem corpus information, such as whether each phrase is a verb, a noun or an adjective, etc., according to the part of speech in the corpus information of the problem.
  • the syntactic structure relevance degree is used to indicate the degree of similarity between the syntactic structure of multiple phrases in the problem corpus information and the syntactic structure of multiple phrases in the problem information.
  • the greater the degree of relevance of the syntactic structure the more likely the syntactic structure of the problem corpus information and the problem information constitute the same, the more likely the problem corpus information and the problem information are the same.
  • the syntactic structure feature vector can be generated according to the syntactic structure of each phrase in the problem corpus information, and the syntactic structure feature vector is generated according to the syntactic structure of each phrase in the problem information, and the syntactic structure feature of the problem corpus information is calculated.
  • the reciprocal of the distance between the vector and the syntactic structure feature vector of the problem information can be used to obtain the syntactic structure relevance.
  • the question information is "what is available funds"
  • the problem corpus information is "where is the available funds?"
  • the syntactic structure formed between the various phrases obtained after the word segmentation is shown in Figure 6, according to the determined syntax.
  • the structure can calculate the syntactic structural similarity of the two.
  • the degree of relevance obtained may be used as the degree of relevance of the problem corpus information to the problem information.
  • statistics may be performed on multiple association degrees to obtain statistical values, such as summation values, weighted summation, or average values. Etc., the obtained statistical value is used as the degree of association between the problem corpus information and the problem information.
  • the above process of obtaining the degree of relevance can regard the degree of relevance as the score of the problem corpus information, and realize the sorting and sorting of the problem corpus information, so that the reply corpus information matching the highest corpus information can be used as the reply information, which can be more accurate. Answer the user's question.
  • a question can have different answers.
  • different responses can be set for multiple users that may appear.
  • the corpus information can be used to determine the matching reply information according to the current user's situation when answering the question for the current user.
  • the corpus information in the professional corpus database includes the problem corpus information and the matched plurality of reply corpus information, and the user status information corresponding to each of the plurality of replies corpus information is different.
  • the server may obtain the user status information of the terminal, and retrieve the problem corpus information matching the problem information from the professional corpus database, determine multiple corpus information corresponding to the problem corpus information, and multiple replies information and user status information.
  • the corresponding reply corpus information is determined as the reply information that matches the question information.
  • multiple states can be divided under the professional category, and the user state information is used to indicate the current state of the user under the professional category.
  • the server may determine the user status information according to the operation performed by the terminal, and store the user identifier of the terminal login with the user status information of the terminal. When the terminal performs the operation of switching the state, the server may also store the user identifier of the terminal login. User status information is updated. For example, in the securities field, the status of the user may include multiple states such as unopened account and account opened. When the terminal performs the account opening operation successfully, the user status information of the terminal is updated from the unopened state to the opened account state.
  • the reply corpus information corresponding to the user state information of the terminal in the multiple corpus information can be considered as the most suitable user.
  • the expected answer, therefore, the reply corpus information corresponding to the user status information is determined as the reply information matching the question information.
  • the corpus information in the professional corpus database includes problem corpus information and matching multiple corpus information, and the user attribute information corresponding to each reply corpus information in the plurality of reply corpus information is different.
  • the server may obtain user attribute information of the terminal, and the user attribute information is used to describe the user of the terminal, retrieve the problem corpus information matching the problem information from the professional corpus database, and determine multiple response corpus information that matches the problem corpus information, The reply corpus information corresponding to the user attribute information among the plurality of reply corpus information is determined as the reply information matching the question information.
  • the user attribute information may include one or more pieces of information such as age, gender, and preference of the user, which is equivalent to a portrait portrayed by the user, and the user attribute information may describe the user, and the user attribute information may be determined according to the user attribute information.
  • Client In the professional corpus database, for different corpus information, different corpus information is set for different user groups, which can simulate the personality and reply mood of different user groups, and the multiple corpus information corresponds to the user attribute information of the terminal.
  • the reply corpus information can be regarded as the answer that best matches the user's expectation, so the reply corpus information corresponding to the user attribute information is determined as the reply information matching the question information.
  • the professional category involved in the embodiment of the present invention may be a specific professional category, and the smart question answering application may solve the problem of the specific professional category for the user, or the professional category may also include multiple professional categories.
  • the smart question-and-answer application can be used as a comprehensive Q&A application to answer questions from various professional categories.
  • the server obtains the reply information matching the problem information based on the question and answer model, and step 306 is performed.
  • the generation mode is used to obtain the reply information, and the generation mode refers to automatically generating the matching reply information by using certain technical means after receiving the input problem information.
  • the question and answer model is used to generate matching reply information for the input question information, and can be obtained by using a machine learning algorithm according to the collected dialog corpus information.
  • the machine learning algorithm can be a Torch deep learning algorithm, or other algorithms.
  • the Torch deep learning algorithm uses a sequence to sequence algorithm model to automatically generate a reply statement based on the input statement. For example, the question information "Are you free tomorrow?" is entered as “Are”, “you", “free”, and “tomorrow” after the word segmentation. After the word segmentation result is input to the sequence to sequence model, the output is obtained. “Yes”, “what's", and “up” constitute the reply message “Yes, what's up”.
  • the server sends a reply message to the terminal.
  • the reply message can be sent to the terminal through the smart question answering application.
  • the reply message is displayed in the question and answer interface.
  • the question information and the reply information can be displayed in the form of a dialogue, simulating the effect of the dialogue between the two users.
  • the terminal can simulate the effect of the user chatting with the customer service personnel, and the problem information is displayed on the right side of the question and answer interface, and the reply information is displayed on the left side of the question and answer interface and on the lower side of the question information.
  • the method provided by the embodiment of the present invention obtains a classification model, and the classification model is used to classify the problem information of the professional category and the problem information of the conversation category.
  • the classification model is used to classify the problem information of the professional category and the problem information of the conversation category.
  • the matching reply information is obtained from the professional corpus database, and is sent to the terminal, which provides a way to reply to the problem information of the professional category, and is not limited to the problem information of the conversation category, and expands the application scope.
  • the user's intention to ask the question can be identified, thereby responding according to the user's intention, improving the accuracy of the reply and improving the intelligence.
  • the relationship between the problem information and the multiple corpus information is calculated from the aspects of semantics, part of speech and syntactic structure, and the corpus information with the greatest relevance to the problem information is determined, and the corpus of the corpus information with the highest degree of relevance is matched.
  • the reply information the information realizes the sorting and sorting of the problem corpus information, and can more accurately answer the user's question.
  • the corresponding reply information is retrieved according to the user state information or the user attribute information, which can satisfy the personalized needs of the user, give an answer that meets the user's expectation, realize differentiated services for different user groups, and improve the pertinence.
  • the matching reply information is obtained based on the classification model, and the manner of replying to the problem information of the conversation category is provided, and the application scope is more comprehensive.
  • the embodiment of the present invention is applied to a scenario in which a user answers a question with a smart question answering application. After the user submits the problem information, the smart question answering application provides the user with the reply information.
  • the user can answer according to the relevant knowledge of the professional category, and carry out related knowledge education and guidance to the user.
  • the user presents the problem information of the conversation category the user can have a chat conversation with the user.
  • the embodiment of the present invention can answer the questions raised by the investor online according to the state of the investor, and combine with the knowledge education function of the investor to answer relevant questions of the securities industry for the investor, prompting Relevant market entry risks provide investors with an environment for systematic learning. At the same time, it is also able to chat with investors to meet the emotional appeal of users.
  • the user inputs the question information “How to open an account” in the question and answer interface. If the user is still in the unopened state at this time, the returned reply information is as shown in FIG. 7 , and the reply information includes not only the account opening process introduction but also the link. The address "go to open an account immediately", the user clicks the link address to initiate the account opening process.
  • the reply message By including a function entry in the reply message, the user can directly trigger the operation through the function entry without having to find the function entry, shortening the user's operation path.
  • the user enters the question information “How to open an account” in the Q&A interface. If the user is in the opened account state at this time, the returned reply message is as shown in FIG. 8 , and the reply message prompts the user that the account has been opened successfully.
  • the user inputs the question information “What is price priority” in the question and answer interface, and the returned reply information is as shown in FIG. 9 , and the concept of “price priority” is explained and exemplified, so that the user can understand the meaning of price priority. It plays a role in imparting securities expertise to users.
  • the user inputs the question information “0XXXX4 (stock code)” in the question and answer interface, and can return the market information of the corresponding stock for the user to view.
  • the user enters the question message "I am signed" in the Q&A interface.
  • the question information is not in the securities field, but is a chat-style statement.
  • the reply message returned at this time is "Congratulations, send a red envelope to celebrate it.” , can chat with the user to meet the user's emotional appeal.
  • the operation process of the entire Q&A method includes an offline process and an online process, where the offline process is at least used to load data into the system memory.
  • the online process is at least used to reply based on data in the system memory.
  • the offline process consists of the following steps:
  • Torch deep learning module uses the Torch deep learning module to train the conversation corpus during the chat, and get the offline question and answer model.
  • the offline indexing module loads the question and answer model and the professional corpus database into the system memory, and the data in the system memory is called when the retrieval module searches online, and the question and answer service is quickly performed.
  • the online process consists of the following steps:
  • the word segmentation module replaces the user's problem with the synonym and then performs the word segmentation, and sends the word segmentation result to the user intention understanding module.
  • the user's intention to understand the module to identify the user's question belongs to the ordinary chat problem or the professional field consultation question.
  • the retrieval module retrieves the candidate answer.
  • the sorting module scores each candidate answer in turn, sorts according to the score, and selects the candidate answer with the highest score.
  • the intelligent question answering method provided by the embodiment of the invention can not only satisfy the user's ordinary chat conversation, emotional conversation, and the like, but also satisfy the user's need for consulting, learning and understanding of the basic knowledge of the professional category.
  • the intelligent question-and-answer method takes into account both the retrieval mode-based reply method and the generative-based reply method, which can provide the existing knowledge base to the user, and can automatically answer the user's chat-style sentence, thereby enhancing the interest. .
  • the intelligent question answering method provided by the embodiment of the invention can effectively replace the manual customer service by using the machine customer service, which greatly saves the number of manual customer service and service time.
  • the machine customer service can provide services to users 24 hours a day, to solve the limitation of the service time of the traditional manual customer service, and the machine customer service does not generate any subjective emotions during the question and answer process, and can provide objective and excellent service under any circumstances. , improved user experience.
  • the above intelligent question and answer method can also be compatible with the manual question and answer mode.
  • the machine customer service and the manual customer service can be allocated reasonably and intelligently, which can enable the user to enjoy the real-time and convenience brought by the machine customer service, and enjoy the realism and personalized service brought by the manual customer service.
  • FIG. 13 is a schematic structural diagram of an intelligent question answering apparatus according to an embodiment of the present invention.
  • the apparatus includes:
  • the receiving module 1301 is configured to receive a Q&A request sent by the terminal.
  • the first reply module 1303 is configured to perform step 304 above;
  • the sending module 1304 is configured to perform step 306 above.
  • the device further includes:
  • the second reply module is configured to perform step 305 above.
  • the first replying module 1303 is further configured to perform the process of determining the reply information according to the degree of association in the foregoing embodiment shown in FIG.
  • the first replying module 1303 is further configured to perform the process of determining the reply information according to the user state information in the embodiment shown in FIG. 3.
  • the first replying module 1303 is further configured to perform the process of determining the reply information according to the user attribute information in the embodiment shown in FIG. 3.
  • FIG. 14 is a schematic structural diagram of an intelligent question answering apparatus according to an embodiment of the present invention.
  • the apparatus includes:
  • the display module 1401 is configured to perform the above step 301 and the above step 307;
  • a determining module 1402 configured to perform the process of determining problem information in step 302 above;
  • the sending module 1403 is configured to perform the process of sending the question and answer request in the foregoing step 302.
  • the display module 1401 is further configured to display the identifier list in the embodiment shown in FIG. 3, and when the selection operation on the public identifier is detected, the question and answer interface in the embodiment shown in FIG. 3 is displayed.
  • the intelligent question answering device provided in the above embodiment is only illustrated by the division of each functional module in the question and answer. In actual applications, the function allocation may be completed by different functional modules as needed. The internal structure of the terminal and the server are divided into different functional modules to perform all or part of the functions described above.
  • the smart question answering device provided by the above embodiment is the same as the smart question answering method embodiment, and the specific implementation process is described in detail in the method embodiment, and details are not described herein again.
  • FIG. 15 is a schematic structural diagram of a terminal according to an embodiment of the present invention.
  • the terminal can be used to implement the functions performed by the terminal in the intelligent question answering method shown in the above embodiments.
  • the terminal includes a processor 1501 and a memory 1502.
  • the memory 1502 stores at least one instruction loaded by the processor 1501 and executed to implement the operations performed by the terminal in the above embodiment.
  • the terminal may further include a receiver or a transmitter to interact with other devices through the receiver or the transmitter.
  • the embodiment of the present invention further provides a computer readable storage medium, where the computer readable storage medium stores at least one instruction loaded by a processor and executed to implement the operations performed by the terminal in the above embodiment.
  • FIG. 16 is a schematic structural diagram of a server according to an embodiment of the present invention.
  • the server includes a processor 1601 and a memory 1602.
  • the memory 1602 stores at least one instruction, and the instruction is loaded and executed by the processor 1601 to implement the foregoing embodiment. The action performed by the server.
  • the server may also include a receiver or transmitter to interact with other devices through the receiver or transmitter.
  • the embodiment of the present invention further provides a computer readable storage medium having at least one instruction stored by a processor and executed to implement the operations performed by the server in the above embodiment.
  • a person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium.
  • the storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.

Abstract

本发明公开了一种智能问答方法、服务器、终端及存储介质，属于互联网技术领域。所述方法包括：接收终端发送的问答请求，问答请求携带问题信息；基于分类模型确定问题信息所属的类别，分类模型用于划分出专业类别的问题信息和对话类别的问题信息；当问题信息属于专业类别时，从专业语料数据库中获取与问题信息匹配的答复信息，专业语料数据库用于存储专业类别的语料信息；向终端发送答复信息，终端用于展示答复信息。本发明提供了对专业类别的问题信息进行答复的方式，扩展了应用范围，并且基于分类模型确定问题信息所属的类别，可以识别用户提出问题的意图，从而根据用户的意图进行答复，提高了答复的精准度，提升了智能化。

Description

智能问答方法、服务器、终端及存储介质
本申请要求于2017年6月9日提交中国国家知识产权局、申请号为201710432438.0、发明名称为“智能问答方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明实施例涉及互联网技术领域，特别涉及一种智能问答方法、服务器、终端及存储介质。
背景技术
智能问答系统是一种新型的信息服务系统，在知识处理、语义识别等功能的基础上能够分析用户意图，快捷精准地为用户解答问题。由于智能问答系统可以代替真人与用户进行对话，而且具有知识面丰富和答复速度快等特点，因此受到了广大用户的喜爱。
为了更好地满足用户的对话需求，智能问答系统可以收集网络用户在闲聊过程中产生的大量对话语料信息，根据收集的对话语料信息训练问答模型。当接收到用户的问题信息时，即可基于该问答模型生成与该问题信息匹配的答复信息，提供给用户。例如，用户输入问题信息“今天天气怎么样”时，基于该问答模型可以为用户提供匹配的答复信息“除了雾霾，都挺好的”。
在实现本发明实施例的过程中，发明人发现相关技术至少存在以下问题：用户提出的问题信息可能包括多种类别，上述问答方法中根据对话语料信息训练的问答模型适用于对话场景，能够为对话类别的问题信息提供匹配的答复信息，但对于其他类别的问题信息则很难提供匹配的答复信息。因此上述问答方法的应用范围存在局限，不够智能。
发明内容
本发明实施例提供了一种智能问答方法、服务器、终端及存储介质，可以解决相关技术的问题。所述技术方案如下：
第一方面，提供了一种智能问答方法，所述方法包括：
接收终端发送的问答请求，所述问答请求携带问题信息；
基于分类模型确定所述问题信息所属的类别，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述专业语料数据库用于存储所述专业类别的语料信息；
向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
第二方面，提供了一种智能问答方法，所述方法包括：
展示智能问答应用的问答界面；
在所述问答界面中，确定待答复的问题信息；
通过所述智能问答应用，向服务器发送问答请求，所述问答请求携带所述问题信息，所述服务器用于基于分类模型确定所述问题信息所属的类别，当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业语料数据库用于存储所述专业类别的语料信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
当接收到所述服务器返回的答复信息时，在所述问答界面中展示所述答复信息。
第三方面，提供了一种智能问答装置，所述装置包括：
接收模块，用于接收终端发送的问答请求，所述问答请求携带问题信息；
分类模块，用于基于分类模型确定所述问题信息所属的类别，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
第一答复模块，用于当所述问题信息属于所述专业类别时，从专业语料数 据库中获取与所述问题信息匹配的答复信息，所述专业语料数据库用于存储所述专业类别的语料信息；
发送模块，用于向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
第四方面，提供了一种智能问答装置，所述装置包括：
展示模块，用于展示智能问答应用的问答界面；
确定模块，用于在所述问答界面中，确定待答复的问题信息；
发送模块，用于通过所述智能问答应用，向服务器发送问答请求，所述问答请求携带所述问题信息，所述服务器用于基于分类模型确定所述问题信息所属的类别，当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业语料数据库用于存储所述专业类别的语料信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
所述展示模块，还用于当接收到所述服务器返回的答复信息时，在所述问答界面中展示所述答复信息。
第五方面，提供了一种服务器，所述服务器包括：处理器和存储器，所述存储器中存储有至少一条指令，所述指令由所述处理器加载并执行以实现如下操作：
接收终端发送的问答请求，所述问答请求携带问题信息；
基于分类模型确定所述问题信息所属的类别，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述专业语料数据库用于存储所述专业类别的语料信息；
向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
第六方面，提供了一种终端，所述终端包括：处理器和存储器，所述存储器中存储有至少一条指令，所述指令由所述处理器加载并执行以实现如下操作：
展示智能问答应用的问答界面；
在所述问答界面中，确定待答复的问题信息；
通过所述智能问答应用，向服务器发送问答请求，所述问答请求携带所述问题信息，所述服务器用于基于分类模型确定所述问题信息所属的类别，当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业语料数据库用于存储所述专业类别的语料信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
当接收到所述服务器返回的答复信息时，在所述问答界面中展示所述答复信息。
第七方面，提供了一种计算机可读存储介质，所述计算机可读存储介质中存储有至少一条指令，所述指令由处理器加载并执行以实现如第一方面所述的智能问答方法中所执行的操作。
第八方面，提供了一种计算机可读存储介质，所述计算机可读存储介质中存储有至少一条指令，所述指令由处理器加载并执行以实现如第二方面所述的智能问答方法中所执行的操作。
本发明实施例提供的技术方案带来的有益效果至少包括：
本发明实施例提供的方法、服务器、终端及存储介质，通过获取分类模型，分类模型用于划分出专业类别的问题信息和对话类别的问题信息，当接收到终端发送的问答请求时，基于分类模型确定问题信息所属的类别，当问题信息属于专业类别时从专业语料数据库中获取匹配的答复信息，发送给终端，提供了对专业类别的问题信息进行答复的方式，扩展了应用范围，并且基于分类模型确定问题信息所属的类别，可以识别用户提出问题的意图，从而根据用户的意 图进行答复，提高了答复的精准度，提升了智能化。
另外，从语义、词性和句法结构等多方面计算问题信息与多个问题语料信息的关联度，确定与问题信息的关联度最大的问题语料信息，将关联度最大的问题语料信息匹配的答复语料信息作为答复信息，实现了对问题语料信息的打分排序，能更加精确地答复用户的问题。
另外，根据用户状态信息或用户属性信息检索对应的答复信息，能够满足用户的个性化需求，给出符合用户预期的答案，实现对不同用户群体的差别化服务，提高了针对性。
另外，当问题信息属于对话类别时基于分类模型获取匹配的答复信息，提供了对对话类别的问题信息进行答复的方式，应用范围更加全面。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的一种实施环境的示意图；
图2是本发明实施例提供的一种实施环境的示意图；
图3是本发明实施例提供的一种智能问答方法的流程图；
图4是本发明实施例提供的一种标识列表的示意图；
图5是本发明实施例提供的一种问答界面的示意图；
图6是本发明实施例提供的一种句法结构的示意图；
图7是本发明实施例提供的一种答复信息的示意图；
图8是本发明实施例提供的一种答复信息的示意图；
图9是本发明实施例提供的一种答复信息的示意图；
图10是本发明实施例提供的一种答复信息的示意图；
图11是本发明实施例提供的一种答复信息的示意图；
图12是本发明实施例提供的一种服务器的架构示意图；
图13是本发明实施例提供的一种智能问答装置的结构示意图；
图14是本发明实施例提供的一种智能问答装置的结构示意图；
图15是本发明实施例提供的一种终端的结构示意图；
图16是本发明实施例提供的一种服务器的结构示意图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
首先，对本发明实施例涉及的一些概念进行如下说明：
1、智能问答应用：用于答复用户提出的问题的应用，可以为在互联网应用中运行的应用，也可以为独立于互联网应用之外的应用客户端。
其中，当智能问答应用在互联网应用中运行时，开发者可以在互联网应用中为智能问答应用注册公共标识，以公共标识来识别该智能问答应用。根据开发者的不同，公共标识可以称为公众号、服务号或者企业号等。在互联网应用中运行智能问答应用后，互联网应用中的普通用户便可关注该智能问答应用的公共标识，或者与该公共标识成为好友关系。之后，互联网应用可以互联网应用中运行的智能问答应用为媒介，为普通用户实现解答问题的功能。
2、互联网应用服务器：是指与互联网应用关联的服务器。
互联网应用服务器提供了开放平台，能够基于原有的互联网应用，向第三方服务器提供接口，通过第三方服务器的接入来增强原有互联网应用的功能或扩展使用该互联网应用资源的平台。互联网应用可以是社交应用、支付应用或者其它相关应用等。
3、智能问答服务器：是指与智能问答应用关联的第三方服务器。
智能问答服务器能够接入到互联网应用服务器，基于互联网应用服务器提供的功能或资源实现解答问题的功能。
图1是本发明实施例提供的一种实施环境的示意图，该实施环境包括：终端101和服务器102，终端101和服务器102之间通过网络连接。
其中，终端101可以为手机、计算机、平板电脑等，服务器102可以是一台服务器，或者由若干台服务器组成的服务器集群，或者是一个云计算服务中心。
终端101用于确定待答复的问题信息，发送给服务器102，服务器102用 于获取该问题信息的答复信息，返回给终端101，从而实现为用户解答问题的功能。
在一种可能的实现方式中，终端101中运行与服务器102关联的智能问答应用，可以通过该智能问答应用与服务器102进行交互。其中，该智能问答应用可以为终端101上安装的应用客户端，也可以为在互联网应用中运行的应用。
在第一种场景下，该智能问答应用为终端101上独立安装的应用客户端，该智能问答应用为与服务器102关联的前端应用，服务器102可以称为智能问答服务器。
在第二种场景下，该智能问答应用为在互联网应用中运行的应用，参见图2，该服务器102可以包括互联网应用服务器1021和智能问答服务器1022，终端101和互联网应用服务器1021之间通过网络连接，互联网应用服务器1021和智能问答服务器1022之间通过网络连接。
互联网应用是与互联网应用服务器1021关联的前端应用，智能问答应用为智能问答服务器1022关联的前端应用。其中，该互联网应用可以为社交应用、支付应用等，相应地，互联网应用服务器可以为社交应用服务器或者支付应用服务器等。
终端101可以安装互联网应用，基于用户标识登录该互联网应用，从而与该互联网应用服务器1021进行交互，该用户标识可以为用户账号、电话号码等。
互联网应用服务器1021用于向智能问答服务器1022(第三方服务器)提供接口，智能问答服务器1022接入互联网应用服务器1021后，智能问答应用服务器1022可在互联网应用服务器1021上注册公共标识，从而基于该公共标识与该互联网应用服务器1021进行交互，并使用该互联网应用服务器1021的资源为用户解答问题。
图3是本发明实施例提供的一种智能问答方法的流程图，该智能问答方法应用于上述实施例所示的实施环境中，交互主体为终端和服务器，参见图3，该方法包括：
301、终端展示智能问答应用的问答界面。
其中，该问答界面可以为HTML5(Hyper Text Markup Language 5，超文本标记语言)形式的页面或者其他形式的页面。
在第一种可能实现方式中，该智能问答应用可以为终端上安装的应用客户端。当用户要解答问题时，在终端上触发启动智能问答应用的操作，终端检测到该启动操作时，启动该智能问答应用，并显示该智能问答应用的问答界面。
在第二种可能实现方式中，终端安装互联网应用，该智能问答应用可以在互联网应用中运行，终端展示互联网应用的界面，在互联网应用的界面中可以启动互联网应用，展示智能问答应用的问答界面。
开发者可以为智能问答应用注册公共标识，该公共标识作为智能问答应用的入口，用于标识该智能问答应用，并且还作为智能问答应用的媒介，连接着用户与智能问答应用。该公共标识可以为智能问答应用的名称、应用编号等。终端可以通过互联网应用并基于登录的用户标识关注该公共标识，则与该用户标识关联的标识列表中即包括该公共标识，当终端通过该互联网应用显示标识列表时，用户可以查看该公共标识，触发对该公共标识的选择操作，该终端检测到对该标识列表中该公共标识的选择操作时，通过智能问答应用展示该问答界面。
其中，终端关注该公共标识时，具体可以通过在互联网应用中搜索该公共标识来关注该公共标识，或者通过在互联网应用中扫描该公共标识的二维码来关注该公共标识。
其中，该标识列表中可以包括该终端关注的一个或多个公共标识，也可以包括该终端的用户关系链中的一个或多个用户标识，如用户的好友、联系人等，还可以包括服务器为该终端提供的服务标识，如天气预报标识等。
参见图4，该标识列表中包括用户的好友以及关注的公共标识，其中的公共标识“微问答”用于为用户解答问题，当用户点击该公共标识“微问答”时，展示如图5所示的问答界面。
302、终端在问答界面中确定待答复的问题信息，通过智能问答应用向服务器发送问答请求。其中，该问答请求携带问题信息，且该问答请求用于请求服务器对问题信息进行答复。
在第一种可能实现方式中，用户可以在该问答界面中输入问题信息，由终端获取输入的问题信息。
问题信息可以包括多种格式，如文本、图片、视频、音频等。对于不同格式的问题信息，其输入方式也不同。例如，问答界面中包括输入栏，用户可以在输入栏中输入文本格式的问题信息；问答界面中还可以包括音频输入按键， 用户在按下音频输入按键时可以输入音频格式的问题信息。
问题信息也可以包括多种类别，如新闻类、计算机技术类、化学类、证券类、对话类等，既可以包括专业类别的问题信息，也可以包括对话类别的问题信息，问题信息的具体类别根据问题信息的内容确定。
在第二种可能实现方式中，终端在问答界面中展示问题信息列表，该问题信息列表包括已设定的多条问题信息，用户可以从中选择要答复的问题信息，由终端获取用户选择的问题信息。其中，该问题信息列表可以由服务器默认设定，或者在对出现次数较多的问题信息进行统计后确定。
当然，也可以将上述两种可能实现方式进行结合以确定问题信息，即终端在问答界面中展示问题信息列表和输入栏，此时用户可以从问题信息列表中选择问题信息，或者也可以在问答界面中输入问题信息。
303、服务器接收到问答请求时，基于分类模型确定问题信息所属的类别。
基于上述步骤301的第一种可能实现方式，该智能问答应用为终端上安装的应用客户端时，该服务器为与智能问答应用关联的智能问答服务器。
基于上述步骤301的第二种可能实现方式，该智能问答应用在互联网应用中运行时，该服务器与互联网应用关联，也与智能问答应用关联。例如，该服务器可以包括与互联网应用关联的互联网应用服务器和与智能问答应用关联的智能问答服务器。
本发明实施例中，服务器可以针对专业类别的问题信息和对话类别的问题信息进行答复。其中，专业类别的问题信息是指问题信息的内容属于专业类别的内容，需要根据专业类别的知识进行答复，该专业类别可以包括计算机技术类、证券类、化学类等一种或多种类别。而对话类别的问题信息是指问题信息的内容属于日常对话内容，不涉及专业类别的知识。
用户可以根据不同的需求确定不同类别的问题信息。例如当用户希望对专业类别的问题进行解答时，输入专业类别的问题信息，当用户希望与智能问答应用闲聊时，输入对话类别的问题信息。相应地，为了满足用户的不同需求，本发明实施例提供了不同的答复方式，当用户提出的问题信息为专业类别的问题信息时，根据专业类别的相关知识为用户提供专业化的答复信息，当用户提出的问题信息为对话类别的问题信息时，与用户进行闲聊式的对话。
因此，当服务器接收到问答请求时，先基于分类模型确定问题信息所属的类别，之后即可按照该问题信息所属的类别对应的答复方式进行答复。
其中，分类模型用于划分出专业类别的问题信息和对话类别的问题信息，根据该分类模型可以确定问题信息是属于专业类别还是属于对话类别。该分类模型可以根据采集的多个样本问题信息及其所属的类别，采用机器学习算法进行训练得到，该机器学习算法可以为支持向量机算法或者其他算法等。
在一种可能实现方式中，在训练分类模型的过程中，可以采集多个样本问题信息及其所属的类别，对每个样本问题信息进行分词后得到每个样本问题信息的多个词组，根据词组的特征构造每个样本问题信息的特征向量，从而根据多个样本问题信息的特征向量和所属的类别训练出该分类模型，该分类模型即可根据问题信息的特征向量确定对应的类别。
相应的，服务器接收到问答请求时，可以先对问题信息进行分词，得到问题信息中的多个词组，根据多个词组中每个词组的特征，构造特征向量，将特征向量输入到分类模型中，即可基于分类模型确定特征向量的类别，该类别即为问题信息所属的类别。
其中，上述分词过程可以基于分词模型进行，该分词模型可以为条件随机场模型、隐马尔科夫模型等。例如，问题信息为“什么是可用资金”，在完成分词之后可以得到多个词组：“什么”、“是”、“可用”、“资金”。
其中，考虑到用户输入问题信息时可能会由于输入错误而导致问题信息中包括错误的词组，则为了提高准确性，服务器可以设置同义词数据库，该同义词数据库中设置了一些词组对应的同义词，这些同义词是词组纠错后得到的正确词组。那么，在对问题信息进行分词之前，服务器可以先对问题信息进行同义词替换，根据同义词数据库，将问题信息中包括的词组替换为对应的同义词，之后再进行分词。例如，用户输入的问题信息为“银行估价多少”，此时可以自动地将问题信息纠正为“银行股价多少”。采用同义词替换的方式可以识别用户的错别字，实现纠错功能，将问题信息定位到正确的语义情境中，以便给出符合用户预期的答复。
另外，考虑到问题信息中包含一些没有实际语义的助词，为了避免这些助词的干扰，以进一步提高答复的准确性，在对问题信息进行分词之后，可以保留具有实际语义的词组，去除掉其他词组，根据保留的词组的特征构造特征向量。
本发明实施例采用分类模型对问题信息进行分类，能够识别用户的意图是要针对专业类别的问题进行咨询还是要进行闲聊，以便根据用户的意图给出用 户满意的答复。进一步地，采用机器学习算法训练分类模型，可以提高准确度，更加精准地理解用户的意图。
304、当问题信息属于专业类别时，服务器从专业语料数据库中获取与问题信息匹配的答复信息，执行步骤306。
当问题信息属于专业类别时，采用检索模式获取答复信息，检索模式是指设置专业语料数据库，当接收到输入的问题信息后，通过在专业语料数据库中以检索匹配的方式进行答复信息的提取。
专业语料数据库用于存储专业类别的语料信息，该语料信息可以包括文本信息、图片信息、视频信息、音频信息和链接信息等多种类型。该专业语料数据库中的语料信息可以采用爬虫下载技术从互联网中获得，也可以根据该智能问答应用提供的功能可能会涉及的语料信息进行设置，或者也可以采用其他方式获得，并且在应用过程中还可以不断地积累用户提出的问题，扩展专业类别的相关知识，并对专业语料数据库进行更新，以使语料信息越来越全面充分，后续能够更加精准地为用户解答问题。在获得语料信息之后，可以采用人工方式进行审核，以确保语料信息的质量。对于其中错误的语料信息，可以采用人工方式进行编辑纠错，以得到正确的语料信息。
例如，以专业类别为证券类别为例，采用爬虫下载技术获得的语料信息可以如下表1所示。
表1
Figure PCTCN2018090422-appb-000001
专业语料数据库中的语料信息可以包括相互匹配的问题语料信息和答复语料信息，也即是可以预先设置问题模板以及对应的答案，作为语料信息存储于专业语料数据库中。那么，服务器确定问题信息属于专业类别时，可以在专业语料数据库中检索出与问题信息匹配的问题语料信息，从而将该问题语料信 息匹配的答复语料信息作为该问题信息的答复信息。
例如，以专业类别为证券类别为例，该语料信息中的问题语料信息和答复语料信息可以如下表2所示。
表2
Figure PCTCN2018090422-appb-000002
如果从专业语料数据库中检索到与问题信息匹配的多个问题语料信息，可以将任一个问题语料信息匹配的答复语料信息作为该问题信息的答复信息，或者，也可以获取多个问题语料信息中每个问题语料信息与问题信息的关联度，关联度表示问题语料信息与问题信息的相似程度，关联度越大表示问题语料信息与问题信息越相似，越可能是相同的问题。因此，确定多个问题语料信息中与问题信息的关联度最大的问题语料信息，即可认为是与该问题信息最为相似的问题语料信息，将关联度最大的问题语料信息匹配的答复语料信息确定为与问题信息匹配的答复信息。
其中，获取问题语料信息与问题信息的关联度时，可以获取问题语料信息与问题信息的语义关联度、词性关联度和句法结构关联度中的至少一项，根据语义关联度、词性关联度和句法结构关联度中的至少一项，获取问题语料信息与问题信息的关联度。
语义关联度用于表示问题语料信息与问题信息在语义上的相似程度，语义关联度越大表示问题语料信息与问题信息的语义越可能相同。计算语义关联度时，可以对问题语料信息和问题信息进行分词，根据问题语料信息中的词组与问题信息中的词组之间的相似度计算语义关联度。其中任两个词组之间的相似度可以为这两个词组的整体特征向量之间的距离的倒数，词组的整体特征向量是指由词组在多个维度上的特征所组成的向量，如由词组的词性特征、语义特征、出现次数特征构成的向量，两个词组的整体特征向量距离越小表示这两个整体特征向量越相似，则这两个词组越相似。例如“沪市”和“上海”的相似度为0.85，“沪市”和“深市”的相似度为0.8，表示“沪市”和“上海”这两个词组更为相似。
词性关联度用于表示问题语料信息中的词组与问题信息的词组在词性上 的相似程度，词性关联度越大表示问题语料信息与问题信息中词组的词性越可能相同，则问题语料信息与问题信息越可能是相同的问题。计算词性关联度时，可以对问题语料信息和问题信息进行分词，获取问题语料信息中各个词组的词性，如各个词组是动词、名词还是形容词等，根据问题语料信息中各个词组的词性构成词性特征向量，并获取问题信息中各个词组的词性，根据问题信息中各个词组的词性构成词性特征向量，通过计算问题语料信息的词性特征向量与问题信息的词性特征向量之间的距离的倒数得到词性关联度。
句法结构关联度用于表示问题语料信息中多个词组构成的句法结构与问题信息中多个词组构成的句法结构的相似程度。句法结构关联度越大表示问题语料信息与问题信息中词组构成的句法结构越可能相同，则问题语料信息与问题信息越可能是相同的问题。计算句法结构关联度时，可以根据问题语料信息中各个词组构成的句法结构生成句法结构特征向量，根据问题信息中各个词组构成的句法结构生成句法结构特征向量，通过计算问题语料信息的句法结构特征向量与问题信息的句法结构特征向量之间的距离的倒数可以得到句法结构关联度。
参见图6，问题信息为“什么是可用资金”，问题语料信息为“可用资金在哪看”，两者分词后得到的各个词组之间构成的句法结构如图6所示，根据确定的句法结构可以计算两者的句法结构相似度。
获取到语义关联度、词性关联度和句法结构关联度中的任一项后，可以将获取的关联度作为问题语料信息与问题信息的关联度。或者，获取到语义关联度、词性关联度和句法结构关联度中的多项后，可以对多项关联度进行统计，得到统计值，如求取和值、进行加权求和或者求取平均值等，将得到的统计值作为问题语料信息与问题信息的关联度。
上述获取关联度的过程，可以将关联度看做问题语料信息的分数，实现了对问题语料信息的打分排序，从而能够将打分最高的问题语料信息匹配的答复语料信息作为答复信息，能够更加精确地答复用户的问题。
考虑到针对不同的用户，一个问题可以有不同的答案，为了更好地满足用户的个性化需求，对于专业语料数据库中的每个问题语料信息，可以针对可能出现的多种用户设置不同的答复语料信息，为当前用户解答问题时可以根据当前用户的情况确定匹配的答复信息。
在一种可能实现方式中，专业语料数据库中的语料信息包括问题语料信息 和匹配的多个答复语料信息，多个答复语料信息中每个答复语料信息对应的用户状态信息不同。则服务器可以获取终端的用户状态信息，并从专业语料数据库中检索与问题信息匹配的问题语料信息，确定与问题语料信息匹配的多个答复语料信息，将多个答复语料信息中与用户状态信息对应的答复语料信息确定为与问题信息匹配的答复信息。
其中，检索与问题信息匹配的问题语料信息时，可以获取与问题信息的关联度最大的问题语料信息，具体过程在此不再赘述。
其中，在该专业类别下可以划分多个状态，用户状态信息用于表示在专业类别下用户当前所处的状态。而服务器可以根据终端执行的操作确定用户状态信息，并将终端登录的用户标识与终端的用户状态信息对应存储，当终端执行了切换状态的操作时，服务器还可以对终端登录的用户标识对应存储的用户状态信息进行更新。例如，在证券领域中，用户的状态可以包括未开户、已开户等多种状态，当终端执行开户操作成功时，将终端的用户状态信息从未开户状态更新为已开户状态。
专业语料数据库中针对一个问题语料信息，为处于不同状态的用户设定了不同的答复语料信息，则多个答复语料信息中与终端的用户状态信息对应的答复语料信息即可认为是最符合用户预期的答案，因此将与用户状态信息对应的答复语料信息确定为与问题信息匹配的答复信息。
在另一种可能实现方式中，专业语料数据库中的语料信息包括问题语料信息和匹配的多个答复语料信息，多个答复语料信息中每个答复语料信息对应的用户属性信息不同。则服务器可以获取终端的用户属性信息，用户属性信息用于对终端的用户进行描述，从专业语料数据库中检索与问题信息匹配的问题语料信息，确定与问题语料信息匹配的多个答复语料信息，将多个答复语料信息中与用户属性信息对应的答复语料信息确定为与问题信息匹配的答复信息。
其中，检索与问题信息匹配的问题语料信息时，可以获取与问题信息的关联度最大的问题语料信息，具体过程在此不再赘述。
其中，该用户属性信息可以包括用户的年龄、性别、喜好等一项或多项信息，相当于为用户刻画的画像，该用户属性信息可以对用户进行描述，根据用户属性信息可以确定用户所属的用户群体。专业语料数据库中针对一个问题语料信息，为不同的用户群体设定了不同的答复语料信息，能够模拟不同用户群组的性格和答复语气，则多个答复语料信息中与终端的用户属性信息对应的答 复语料信息即可认为是最符合用户预期的答案，因此将与用户属性信息对应的答复语料信息确定为与问题信息匹配的答复信息。
需要说明的是，本发明实施例中涉及的专业类别可以为某一种特定的专业类别，则该智能问答应用可以为用户解答特定专业类别的问题，或者该专业类别也可以包括多种专业类别，则该智能问答应用可以作为一个综合型问答应用，为用户解答各种专业类别的问题。
305、当问题信息属于对话类别时，服务器基于问答模型获取问题信息匹配的答复信息，执行步骤306。
当问题信息属于对话类别时，采用生成模式获取答复信息，生成模式是指在接收到输入的问题信息后采用一定的技术手段自动生成匹配的答复信息。
其中，该问答模型用于为输入的问题信息生成匹配的答复信息，可以根据采集的对话语料信息，采用机器学习算法进行训练得到。该机器学习算法可以为Torch深度学习算法，或者其他算法等。
Torch深度学习算法采用sequence to sequence(语句到语句)的算法模型自动地根据输入语句生成答复语句。例如，输入的问题信息“Are you free tomorrow？”在进行分词以后就变成“Are”、“you”、“free”、“tomorrow”，在将分词结果输入到sequence to sequence模型后，得到输出“Yes”、“what’s”、“up”，构成答复信息“Yes，what’s up”。
306、服务器向终端发送答复信息。
当服务器执行上述步骤304或者步骤305获取到答复信息时，可以通过该智能问答应用向终端发送答复信息。
307、终端接收到答复信息时，在问答界面中展示答复信息。
终端展示答复信息时，可以将该问题信息与该答复信息以对话的形式进行展示，模拟两个用户之间进行对话的效果。例如，终端可以模拟用户与客服人员聊天的效果，将问题信息在问答界面的右侧展示，将答复信息在问答界面的左侧、问题信息的下侧展示。
本发明实施例提供的方法，通过获取分类模型，分类模型用于划分出专业类别的问题信息和对话类别的问题信息，当接收到终端发送的问答请求时，基于分类模型确定问题信息所属的类别，当问题信息属于专业类别时从专业语料数据库中获取匹配的答复信息，发送给终端，提供了对专业类别的问题信息进行答复的方式，而不仅限于对话类别的问题信息，扩展了应用范围，并且基于 分类模型确定问题信息所属的类别，可以识别用户提出问题的意图，从而根据用户的意图进行答复，提高了答复的精准度，提升了智能化。
另外，从语义、词性和句法结构等多方面计算问题信息与多个问题语料信息的关联度，确定与问题信息的关联度最大的问题语料信息，将关联度最大的问题语料信息匹配的答复语料信息作为答复信息，实现了对问题语料信息的打分排序，能更加精确地答复用户的问题。
另外，根据用户状态信息或用户属性信息检索对应的答复信息，能够满足用户的个性化需求，给出符合用户预期的答案，实现对不同用户群体的差别化服务，提高了针对性。
另外，当问题信息属于对话类别时基于分类模型获取匹配的答复信息，提供了对对话类别的问题信息进行答复的方式，应用范围更加全面。
本发明实施例应用于用户与智能问答应用进行问答的场景下，用户提出问题信息后，由智能问答应用为用户提供答复信息。
在用户提出专业类别的问题信息时，可以根据专业类别的相关知识进行解答，对用户进行相关的知识教育和指导。而在用户提出对话类别的问题信息时，可以与用户进行闲聊式的对话。
其中，专业类别为证券类别时，本发明实施例可以根据投资者的状态线上答复投资者提出的问题，并与投资者的知识教育功能相结合，为投资者解答证券行业的相关问题，提示相关入市风险，为投资者提供了一个系统学习知识的环境。同时还能够与投资者进行闲聊式的对话，满足用户的情感诉求。
例如，用户在问答界面中输入问题信息“怎么开户”，如果此时用户还处于未开户的状态，则返回的答复信息如图7所示，该答复信息中不仅包括开户流程介绍，还包括链接地址“立即去开户”，用户点击该链接地址即可发起开户流程。通过在答复信息中包含功能入口，使用户直接通过功能入口触发操作，而无需查找该功能入口，缩短了用户的操作路径。
用户在问答界面中输入问题信息“怎么开户”，如果此时用户处于已开户的状态，则返回的答复信息如图8所示，该答复信息提示用户已经开户成功。
参见图9，用户在问答界面中输入问题信息“什么是价格优先”，返回的答复信息如图9所示，对“价格优先”的概念进行了说明和举例，以便用户了解价格优先的含义，起到了为用户传授证券类专业知识的作用。
参见图10，用户在问答界面中输入问题信息“0XXXX4(股票代码)”，可以返回相应股票的行情信息，供用户查看。
参见图11，用户在问答界面中输入问题信息“我中签啦”，该问题信息不属于证券领域，而是闲聊式的语句，此时返回的答复信息为“恭喜，发个红包庆祝一下吧”，能够与用户闲聊，满足用户的情感诉求。
图12是本发明实施例提供的一种服务器的架构示意图，基于图12所示的架构，整个问答方法的操作流程包括离线流程和在线流程，其中离线流程至少用于加载数据到系统内存中，在线流程至少用于根据系统内存中的数据进行答复。
离线流程包括以下步骤：
1、采用爬虫下载技术或者数据积累构建专业语料数据库，包括专业模板，专业词汇，专业同义词等。
2、利用Torch深度学习模块对闲聊时的对话语料进行训练，得到离线的问答模型。
3、离线索引模块将问答模型和专业语料数据库加载到系统内存中，供检索模块在线检索时调用系统内存中的数据，快速地进行问答服务。
在线流程包括以下步骤：
1、接收用户的问答请求，发送给分词模块。
2、分词模块将用户的问题进行同义词替换后再进行分词，将分词结果发送给用户意图理解模块，由用户意图理解模块识别用户的问题属于普通闲聊的问题还是专业领域咨询的问题。
3、确定用户的意图之后，检索模块检索出候选答案。
4、排序模块依次对每个候选答案进行打分，按照打分进行排序，选择分数最高的候选答案。
本发明实施例提供的智能问答方法，整体上既能够满足用户普通的闲聊时聊天对话、情感倾诉等诉求，又能够满足用户对于专业类别的基础知识的咨询、学习和了解的需求。该智能问答方法同时兼顾了基于检索模式的答复方式和基于生成式的答复方式，既能够将现有的知识库提供给用户，也能够对用户的闲聊式语句进行自动地回答，增强了趣味性。
采用本发明实施例提供的智能问答方法可以使用机器客服有效地替代人工客服，大量节约了人工客服的数量和服务时间。而且，机器客服可以全天候24小时地为用户提供服务，解决了传统人工客服的服务时间的限制，并且机器客服在问答过程中不会产生任何主观情绪，在任何情况下均可提供客观优质的服务，提升了用户体验。
当然，上述智能问答方法也可以与人工问答的方式兼容，根据用户数量和用户特性，合理、智能地分配机器客服和人工客服，既能够让用户享受机器客服带来的实时性和便捷性，又能享受人工客服带来的真实感和个性化的服务。
图13是本发明实施例提供的一种智能问答装置的结构示意图。参见图13，该装置包括：
接收模块1301，用于接收终端发送的问答请求；
分类模块1302，用于执行上述步骤303中的分类过程；
第一答复模块1303，用于执行上述步骤304；
发送模块1304，用于执行上述步骤306。
可选地，装置还包括：
第二答复模块，用于执行上述步骤305。
可选地，第一答复模块1303还用于执行上述图3所示实施例中根据关联度确定答复信息的过程。
可选地，第一答复模块1303还用于执行上述图3所示实施例中根据用户状态信息确定答复信息的过程。
可选地，第一答复模块1303还用于执行上述图3所示实施例中根据用户属性信息确定答复信息的过程。
图14是本发明实施例提供的一种智能问答装置的结构示意图。参见图14，该装置包括：
展示模块1401，用于执行上述步骤301和上述步骤307；
确定模块1402，用于执行上述步骤302中确定问题信息的过程；
发送模块1403，用于执行上述步骤302中发送问答请求的过程。
可选地，展示模块1401，还用于展示上述图3所示实施例中的标识列表，并在检测到对公共标识的选择操作时，展示上述图3所示实施例中的问答界面。
需要说明的是：上述实施例提供的智能问答装置在进行问答时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成，即将终端和服务器的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。另外，上述实施例提供的智能问答装置与智能问答方法实施例属于同一构思，其具体实现过程详见方法实施例，这里不再赘述。
图15是本发明实施例提供的一种终端的结构示意图。该终端可以用于实施上述实施例所示出的智能问答方法中的终端所执行的功能。
该终端包括处理器1501和存储器1502，存储器1502中存储有至少一条指令，该指令由处理器1501加载并执行以实现上述实施例中终端所执行的操作。
可选地，该终端还可以包括接收器或发射器，通过接收器或发射器与其他设备进行交互。
本发明实施例还提供了一种计算机可读存储介质，该计算机可读存储介质中存储有至少一条指令，该指令由处理器加载并执行以实现上述实施例中终端所执行的操作。
图16是本发明实施例提供的一种服务器的结构示意图，该服务器包括处理器1601和存储器1602，存储器1602中存储有至少一条指令，该指令由处理器1601加载并执行以实现上述实施例中服务器所执行的操作。
可选地，该服务器还可以包括接收器或发射器，通过接收器或发射器与其他设备进行交互。
本发明实施例还提供了一种计算机可读存储介质，该计算机可读存储介质中存储有至少一条指令，该指令由处理器加载并执行以实现上述实施例中服务器所执行的操作。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成，也可以通过程序来指令相关的硬件完成，所述的程序可以存储于一种计算机可读存储介质中，上述提到的存储介质可以是只读存储器，磁盘或光盘等。
以上所述仅为本发明的可选实施例，并不用以限制本发明实施例，凡在本发明实施例的精神和原则之内，所作的任何修改、等同替换、改进等，均应包 含在本发明实施例的保护范围之内。

Claims (24)

  1. 一种智能问答方法，其特征在于，应用于服务器中，所述方法包括：
    接收终端发送的问答请求，所述问答请求携带问题信息；
    基于分类模型确定所述问题信息所属的类别，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
    当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述专业语料数据库用于存储所述专业类别的语料信息；
    向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
  2. 根据权利要求1所述的方法，其特征在于，所述基于分类模型确定所述问题信息所属的类别，包括：
    对所述问题信息进行分词，得到所述问题信息中的多个词组；
    根据所述多个词组中每个词组的特征，构造特征向量；
    将所述特征向量输入到所述分类模型中，基于所述分类模型确定所述特征向量的类别，作为所述问题信息所属的类别。
  3. 根据权利要求1或2所述的方法，其特征在于，所述语料信息包括相互匹配的问题语料信息和答复语料信息，所述从专业语料数据库中获取与所述问题信息匹配的答复信息，包括：
    从所述专业语料数据库中检索与所述问题信息匹配的多个问题语料信息；
    获取所述多个问题语料信息中每个问题语料信息与所述问题信息的关联度；
    确定所述多个问题语料信息中与所述问题信息的关联度最大的问题语料信息；
    将所述关联度最大的问题语料信息匹配的答复语料信息确定为与所述问题信息匹配的答复信息。
  4. 根据权利要求3所述的方法，其特征在于，所述获取所述多个问题语料信息中每个问题语料信息与所述问题信息的关联度，包括：
    对于每个问题语料信息，获取所述问题语料信息与所述问题信息的语义关联度、词性关联度和句法结构关联度中的至少一项；
    根据所述语义关联度、所述词性关联度和所述句法结构关联度中的至少一项，获取所述问题语料信息与所述问题信息的关联度。
  5. 根据权利要求1或2所述的方法，其特征在于，所述语料信息包括问题语料信息和匹配的多个答复语料信息，所述多个答复语料信息中每个答复语料信息对应的用户状态信息不同；
    所述从专业语料数据库中获取与所述问题信息匹配的答复信息，包括：
    获取所述终端的用户状态信息，所述用户状态信息用于表示在所述专业类别下所述终端的用户当前所处的状态；
    从所述专业语料数据库中检索与所述问题信息匹配的问题语料信息，确定与所述问题语料信息匹配的多个答复语料信息；
    将所述多个答复语料信息中与所述用户状态信息对应的答复语料信息确定为与所述问题信息匹配的答复信息。
  6. 根据权利要求1或2所述的方法，其特征在于，所述语料信息包括问题语料信息和匹配的多个答复语料信息，所述多个答复语料信息中每个答复语料信息对应的用户属性信息不同；
    所述从专业语料数据库中获取与所述问题信息匹配的答复信息，包括：
    获取所述终端的用户属性信息，所述用户属性信息用于对所述终端的用户进行描述；
    从所述专业语料数据库中检索与所述问题信息匹配的问题语料信息，确定与所述问题语料信息匹配的多个答复语料信息；
    将所述多个答复语料信息中与所述用户属性信息对应的答复语料信息确定为与所述问题信息匹配的答复信息。
  7. 根据权利要求1所述的方法，其特征在于，所述基于分类模型确定所述问题信息所属的类别之后，所述方法还包括：
    当所述问题信息属于所述对话类别时，基于问答模型获取所述问题信息匹配的答复信息，所述问答模型根据收集的对话语料信息训练得到；
    向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
  8. 一种智能问答方法，其特征在于，应用于终端中，所述方法包括：
    展示智能问答应用的问答界面；
    在所述问答界面中，确定待答复的问题信息；
    通过所述智能问答应用，向服务器发送问答请求，所述问答请求携带所述问题信息，所述服务器用于基于分类模型确定所述问题信息所属的类别，当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业语料数据库用于存储所述专业类别的语料信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
    当接收到所述服务器返回的答复信息时，在所述问答界面中展示所述答复信息。
  9. 根据权利要求8所述的方法，其特征在于，所述智能问答应用在互联网应用中运行，所述展示智能问答应用的问答界面，包括：
    通过所述互联网应用显示与登录的用户标识关联的标识列表，所述标识列表中包括所述智能问答应用的公共标识；
    当检测到对所述公共标识的选择操作时，通过所述智能问答应用展示所述问答界面。
  10. 一种智能问答装置，其特征在于，应用于服务器中，所述装置包括：
    接收模块，用于接收终端发送的问答请求，所述问答请求携带问题信息；
    分类模块，用于基于分类模型确定所述问题信息所属的类别，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
    第一答复模块，用于当所述问题信息属于所述专业类别时，从专业语料数 据库中获取与所述问题信息匹配的答复信息，所述专业语料数据库用于存储所述专业类别的语料信息；
    发送模块，用于向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
  11. 根据权利要求10所述的装置，其特征在于，所述语料信息包括问题语料信息和匹配的多个答复语料信息，所述多个答复语料信息中每个答复语料信息对应的用户状态信息不同；
    所述第一答复模块用于获取所述终端的用户状态信息，所述用户状态信息用于表示在所述专业类别下所述终端的用户当前所处的状态；从所述专业语料数据库中检索与所述问题信息匹配的问题语料信息，确定与所述问题语料信息匹配的多个答复语料信息；将所述多个答复语料信息中与所述用户状态信息对应的答复语料信息确定为与所述问题信息匹配的答复信息。
  12. 根据权利要求10所述的装置，其特征在于，所述语料信息包括问题语料信息和匹配的多个答复语料信息，所述多个答复语料信息中每个答复语料信息对应的用户属性信息不同；
    所述第一答复模块用于获取所述终端的用户属性信息，所述用户属性信息用于对所述终端的用户进行描述；从所述专业语料数据库中检索与所述问题信息匹配的问题语料信息，确定与所述问题语料信息匹配的多个答复语料信息；将所述多个答复语料信息中与所述用户属性信息对应的答复语料信息确定为与所述问题信息匹配的答复信息。
  13. 一种智能问答装置，其特征在于，应用于终端中，所述装置包括：
    展示模块，用于展示智能问答应用的问答界面；
    确定模块，用于在所述问答界面中，确定待答复的问题信息；
    发送模块，用于通过所述智能问答应用，向服务器发送问答请求，所述问答请求携带所述问题信息，所述服务器用于基于分类模型确定所述问题信息所属的类别，当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业语料数据库用于存储所述专业类别的语料信 息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
    所述展示模块，还用于当接收到所述服务器返回的答复信息时，在所述问答界面中展示所述答复信息。
  14. 一种服务器，其特征在于，所述服务器包括：处理器和存储器，所述存储器中存储有至少一条指令，所述指令由所述处理器加载并执行以实现如下操作：
    接收终端发送的问答请求，所述问答请求携带问题信息；
    基于分类模型确定所述问题信息所属的类别，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
    当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述专业语料数据库用于存储所述专业类别的语料信息；
    向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
  15. 根据权利要求14所述的服务器，其特征在于，所述指令由所述处理器加载并执行以实现如下操作：
    对所述问题信息进行分词，得到所述问题信息中的多个词组；
    根据所述多个词组中每个词组的特征，构造特征向量；
    将所述特征向量输入到所述分类模型中，基于所述分类模型确定所述特征向量的类别，作为所述问题信息所属的类别。
  16. 根据权利要求14或15所述的服务器，其特征在于，所述指令由所述处理器加载并执行以实现如下操作：
    从所述专业语料数据库中检索与所述问题信息匹配的多个问题语料信息；
    获取所述多个问题语料信息中每个问题语料信息与所述问题信息的关联度；
    确定所述多个问题语料信息中与所述问题信息的关联度最大的问题语料信息；
    将所述关联度最大的问题语料信息匹配的答复语料信息确定为与所述问题信息匹配的答复信息。
  17. 根据权利要求16所述的服务器，其特征在于，所述指令由所述处理器加载并执行以实现如下操作：
    对于每个问题语料信息，获取所述问题语料信息与所述问题信息的语义关联度、词性关联度和句法结构关联度中的至少一项；
    根据所述语义关联度、所述词性关联度和所述句法结构关联度中的至少一项，获取所述问题语料信息与所述问题信息的关联度。
  18. 根据权利要求14或15所述的服务器，其特征在于，所述语料信息包括问题语料信息和匹配的多个答复语料信息，所述多个答复语料信息中每个答复语料信息对应的用户状态信息不同；所述指令由所述处理器加载并执行以实现如下操作：
    获取所述终端的用户状态信息，所述用户状态信息用于表示在所述专业类别下所述终端的用户当前所处的状态；
    从所述专业语料数据库中检索与所述问题信息匹配的问题语料信息，确定与所述问题语料信息匹配的多个答复语料信息；
    将所述多个答复语料信息中与所述用户状态信息对应的答复语料信息确定为与所述问题信息匹配的答复信息。
  19. 根据权利要求14或15所述的服务器，其特征在于，所述语料信息包括问题语料信息和匹配的多个答复语料信息，所述多个答复语料信息中每个答复语料信息对应的用户属性信息不同；所述指令由所述处理器加载并执行以实现如下操作：
    获取所述终端的用户属性信息，所述用户属性信息用于对所述终端的用户进行描述；
    从所述专业语料数据库中检索与所述问题信息匹配的问题语料信息，确定与所述问题语料信息匹配的多个答复语料信息；
    将所述多个答复语料信息中与所述用户属性信息对应的答复语料信息确定为与所述问题信息匹配的答复信息。
  20. 根据权利要求14所述的服务器，其特征在于，所述指令由所述处理器加载并执行以实现如下操作：
    当所述问题信息属于所述对话类别时，基于问答模型获取所述问题信息匹配的答复信息，所述问答模型根据收集的对话语料信息训练得到；
    向所述终端发送所述答复信息，所述终端用于展示所述答复信息。
  21. 一种终端，其特征在于，所述终端包括：处理器和存储器，所述存储器中存储有至少一条指令，所述指令由所述处理器加载并执行以实现如下操作：
    展示智能问答应用的问答界面；
    在所述问答界面中，确定待答复的问题信息；
    通过所述智能问答应用，向服务器发送问答请求，所述问答请求携带所述问题信息，所述服务器用于基于分类模型确定所述问题信息所属的类别，当所述问题信息属于所述专业类别时，从专业语料数据库中获取与所述问题信息匹配的答复信息，所述分类模型用于划分出专业类别的问题信息和对话类别的问题信息，所述专业语料数据库用于存储所述专业类别的语料信息，所述专业类别的问题信息是指所述问题信息的内容涉及所述专业类别的知识，所述对话类别的问题信息是指所述问题信息的内容不涉及所述专业类别的知识；
    当接收到所述服务器返回的答复信息时，在所述问答界面中展示所述答复信息。
  22. 根据权利要求21所述的终端，其特征在于，所述智能问答应用在互联网应用中运行，所述指令由所述处理器加载并执行以实现如下操作：
    通过所述互联网应用显示与登录的用户标识关联的标识列表，所述标识列表中包括所述智能问答应用的公共标识；
    当检测到对所述公共标识的选择操作时，通过所述智能问答应用展示所述问答界面。
  23. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质中 存储有至少一条指令，所述指令由处理器加载并执行以实现如权利要求1至权利要求7任一项所述的智能问答方法中所执行的操作。
  24. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质中存储有至少一条指令，所述指令由处理器加载并执行以实现如权利要求8或权利要求9所述的智能问答方法中所执行的操作。
PCT/CN2018/090422 2017-06-09 2018-06-08 智能问答方法、服务器、终端及存储介质 WO2018224034A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710432438.0 2017-06-09
CN201710432438.0A CN107301213A (zh) 2017-06-09 2017-06-09 智能问答方法及装置

Publications (1)

Publication Number Publication Date
WO2018224034A1 true WO2018224034A1 (zh) 2018-12-13

Family

ID=60135724

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/090422 WO2018224034A1 (zh) 2017-06-09 2018-06-08 智能问答方法、服务器、终端及存储介质

Country Status (2)

Country Link
CN (1) CN107301213A (zh)
WO (1) WO2018224034A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112925890A (zh) * 2021-03-05 2021-06-08 湖南神通智能股份有限公司 一种智能问答系统
CN113094474A (zh) * 2021-06-08 2021-07-09 深圳追一科技有限公司 智能问答方法和装置、服务器、计算机可读存储介质

Families Citing this family (49)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107301213A (zh) * 2017-06-09 2017-10-27 腾讯科技（深圳）有限公司 智能问答方法及装置
CN108415983B (zh) * 2018-02-10 2022-04-15 武汉市灯塔互动文化传播有限公司 一种基于交互的智能问题解决方法和装置
CN108491382A (zh) * 2018-03-14 2018-09-04 四川大学 一种半监督生物医学文本语义消歧方法
CN108595494B (zh) * 2018-03-15 2022-05-20 腾讯科技（深圳）有限公司 答复信息的获取方法及装置
CN110019741B (zh) * 2018-06-01 2023-11-14 中国平安人寿保险股份有限公司 问答系统答案匹配方法、装置、设备及可读存储介质
CN109947909B (zh) * 2018-06-19 2024-03-12 平安科技（深圳）有限公司 智能客服应答方法、设备、存储介质及装置
CN108491394A (zh) * 2018-06-27 2018-09-04 杭州贝店科技有限公司 一种语义分析方法、装置、计算机设备及存储介质
CN110019316A (zh) * 2018-07-06 2019-07-16 新疆玖富万卡信息技术有限公司 信息交互方法和服务器
CN109063152A (zh) * 2018-08-08 2018-12-21 鲸数科技（北京）有限公司 智能问答方法、装置及智能终端
CN110162675B (zh) * 2018-09-25 2023-05-02 腾讯科技（深圳）有限公司 应答语句的生成方法、装置、计算机可读介质及电子设备
WO2020062006A1 (en) * 2018-09-28 2020-04-02 Entit Software Llc Intent and context-aware dialogue based virtual assistance
CN111046151B (zh) * 2018-10-15 2023-04-21 腾讯科技（深圳）有限公司 一种消息处理方法及装置
CN109451188B (zh) * 2018-11-29 2022-03-18 平安科技（深圳）有限公司 差异性自助应答的方法、装置、计算机设备和存储介质
CN109598658B (zh) * 2018-11-30 2023-05-02 暨南大学 一种新生入学服务系统和方法
CN111382234A (zh) * 2018-12-11 2020-07-07 航天信息股份有限公司 一种基于客服的答复提供方法及装置
CN109829038A (zh) * 2018-12-11 2019-05-31 平安科技（深圳）有限公司 基于深度学习的问答反馈方法、装置、设备及存储介质
CN109658271A (zh) * 2018-12-19 2019-04-19 前海企保科技（深圳）有限公司 一种基于保险专业场景的智能客服系统及方法
CN109815321B (zh) * 2018-12-26 2020-12-11 出门问问信息科技有限公司 问答方法、装置、设备及存储介质
CN109829044A (zh) * 2018-12-28 2019-05-31 北京百度网讯科技有限公司 对话方法、装置及设备
CN109859747B (zh) * 2018-12-29 2021-05-04 北京百度网讯科技有限公司 语音交互方法、设备以及存储介质
CN109753561B (zh) * 2019-01-16 2021-04-27 长安汽车金融有限公司 一种自动回复的生成方法及装置
CN109885666A (zh) * 2019-01-18 2019-06-14 科大国创软件股份有限公司 一种基于html5的智能语音客服机器人的方法及系统
CN111666006B (zh) * 2019-03-05 2022-01-14 京东方科技集团股份有限公司 画作问答方法及装置、画作问答系统、可读存储介质
CN110321414B (zh) * 2019-04-19 2023-04-28 四川政资汇智能科技有限公司 一种基于深度学习的人工智能咨询服务方法及系统
CN110134775B (zh) * 2019-05-10 2021-08-24 中国联合网络通信集团有限公司 问答数据生成方法及装置、存储介质
CN110516063A (zh) * 2019-07-11 2019-11-29 网宿科技股份有限公司 一种服务系统的更新方法、电子设备及可读存储介质
CN111881266A (zh) * 2019-07-19 2020-11-03 马上消费金融股份有限公司 一种应答方法及装置
CN110428267B (zh) * 2019-07-31 2023-01-06 联想(北京)有限公司 内容处理方法及装置、计算机系统以及可读存储介质
CN110516054A (zh) * 2019-08-15 2019-11-29 出门问问(武汉)信息科技有限公司 一种垃圾分类问答方法、设备以及可读存储介质
CN110472033A (zh) * 2019-08-16 2019-11-19 北京一链数云科技有限公司 基于nlp模型的问答方法、装置及服务器
CN110807325B (zh) * 2019-10-18 2023-05-26 腾讯科技（深圳）有限公司 谓词识别方法、装置及存储介质
CN110941708B (zh) * 2019-11-04 2023-02-28 智器云南京信息科技有限公司 智能问答库建立方法、智能问答方法及装置、计算机设备
CN111159015B (zh) * 2019-12-13 2022-01-14 华为技术有限公司 定位问题的方法和装置
CN111104503A (zh) * 2019-12-24 2020-05-05 华中科技大学 一种建筑工程质量验收规范问答系统及其构建方法
CN111241833A (zh) * 2020-01-16 2020-06-05 支付宝(杭州)信息技术有限公司 一种文本数据的分词方法、装置及电子设备
CN111476669A (zh) * 2020-03-26 2020-07-31 杭州十尾网络科技有限公司 数据分析方法和装置
CN111292103A (zh) * 2020-04-24 2020-06-16 支付宝(杭州)信息技术有限公司 向用户推荐客服渠道的方法和装置
CN111581950B (zh) * 2020-04-30 2024-01-02 支付宝(杭州)信息技术有限公司 同义名称词的确定方法和同义名称词的知识库的建立方法
CN111831810B (zh) * 2020-07-23 2024-02-09 中国平安人寿保险股份有限公司 智能问答方法、装置、设备及存储介质
CN112214654A (zh) * 2020-10-19 2021-01-12 厦门渊亭信息科技有限公司 一种通用的智能问答自动化运维系统及方法
CN112328871B (zh) * 2020-10-27 2024-04-26 深圳集智数字科技有限公司 一种基于rpa模块的回复生成方法、装置、设备及存储介质
CN112085422B (zh) * 2020-10-28 2021-06-22 杭州环研科技有限公司 一种基于人工智能的环保在线服务系统
CN112597287B (zh) * 2020-12-15 2024-03-15 深圳市优必选科技股份有限公司 一种语句处理方法、语句处理装置及智能设备
CN112992137B (zh) * 2021-01-29 2022-12-06 青岛海尔科技有限公司 语音交互方法和装置、存储介质及电子装置
CN113010654A (zh) * 2021-03-17 2021-06-22 北京十一贝科技有限公司 应用于保险行业的问题回复方法、装置、电子设备和介质
CN113377934B (zh) * 2021-05-21 2022-07-05 海南师范大学 一种实现智能客服的系统及方法
CN113343713B (zh) * 2021-06-30 2022-06-17 中国平安人寿保险股份有限公司 意图识别方法、装置、计算机设备及存储介质
WO2024036616A1 (zh) * 2022-08-19 2024-02-22 华为技术有限公司 一种基于终端的问答方法及装置
CN115617973B (zh) * 2022-12-14 2023-03-21 安徽数分智能科技有限公司 一种基于智能数据处理的信息获取方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105630938A (zh) * 2015-12-23 2016-06-01 深圳市智客网络科技有限公司 一种智能问答系统
CN105893465A (zh) * 2016-03-28 2016-08-24 北京京东尚科信息技术有限公司 自动问答方法和装置
CN106649561A (zh) * 2016-11-10 2017-05-10 复旦大学 面向税务咨询业务的智能问答系统
CN106682137A (zh) * 2016-12-19 2017-05-17 武汉市灯塔互动文化传播有限公司 一种智能股票投顾问答交互方法与系统
CN106708924A (zh) * 2016-11-09 2017-05-24 上海知邦信息科技有限公司 一种咨询系统和方法
CN107301213A (zh) * 2017-06-09 2017-10-27 腾讯科技（深圳）有限公司 智能问答方法及装置

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101076061A (zh) * 2007-03-30 2007-11-21 腾讯科技（深圳）有限公司 一种机器人服务器及自动聊天方法
CN102141997A (zh) * 2010-02-02 2011-08-03 三星电子（中国）研发中心 智能决策支持系统及其智能决策方法
CN104598445B (zh) * 2013-11-01 2019-05-10 腾讯科技（深圳）有限公司 自动问答系统和方法
CN103902652A (zh) * 2014-02-27 2014-07-02 深圳市智搜信息技术有限公司 自动问答系统
CN105159996B (zh) * 2015-09-07 2018-09-07 百度在线网络技术（北京）有限公司 基于人工智能的深度问答服务提供方法和装置
CN106789595A (zh) * 2017-01-17 2017-05-31 北京诸葛找房信息技术有限公司 信息推送方法和装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105630938A (zh) * 2015-12-23 2016-06-01 深圳市智客网络科技有限公司 一种智能问答系统
CN105893465A (zh) * 2016-03-28 2016-08-24 北京京东尚科信息技术有限公司 自动问答方法和装置
CN106708924A (zh) * 2016-11-09 2017-05-24 上海知邦信息科技有限公司 一种咨询系统和方法
CN106649561A (zh) * 2016-11-10 2017-05-10 复旦大学 面向税务咨询业务的智能问答系统
CN106682137A (zh) * 2016-12-19 2017-05-17 武汉市灯塔互动文化传播有限公司 一种智能股票投顾问答交互方法与系统
CN107301213A (zh) * 2017-06-09 2017-10-27 腾讯科技（深圳）有限公司 智能问答方法及装置

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112925890A (zh) * 2021-03-05 2021-06-08 湖南神通智能股份有限公司 一种智能问答系统
CN113094474A (zh) * 2021-06-08 2021-07-09 深圳追一科技有限公司 智能问答方法和装置、服务器、计算机可读存储介质
CN113094474B (zh) * 2021-06-08 2022-05-10 深圳追一科技有限公司 智能问答方法和装置、服务器、计算机可读存储介质

Also Published As

Publication number Publication date
CN107301213A (zh) 2017-10-27

Similar Documents

Publication Publication Date Title
WO2018224034A1 (zh) 智能问答方法、服务器、终端及存储介质
CN113128239B (zh) 促进以多种语言与自动化助理的端到端沟通
US11599729B2 (en) Method and apparatus for intelligent automated chatting
CN107886949B (zh) 一种内容推荐方法及装置
US9582757B1 (en) Scalable curation system
US20200137001A1 (en) Generating responses in automated chatting
WO2022111244A1 (zh) 数据处理方法、装置、电子设备及存储介质
US9710829B1 (en) Methods, systems, and articles of manufacture for analyzing social media with trained intelligent systems to enhance direct marketing opportunities
JP2017534941A (ja) オーファン発話検出システム及び方法
CN112800218A (zh) 基于消息分类自动扩增消息交换话题
CN110399470B (zh) 会话消息处理
WO2020253064A1 (zh) 语音的识别方法及装置、计算机设备、存储介质
CN112699645B (zh) 语料标注方法、装置及设备
CN110018823B (zh) 交互式应用程序的处理方法及系统、生成方法及系统
CN116127020A (zh) 生成式大语言模型训练方法以及基于模型的搜索方法
CN112417158A (zh) 文本数据分类模型的训练方法、分类方法、装置和设备
CN116226334A (zh) 生成式大语言模型训练方法以及基于模型的搜索方法
CN111639162A (zh) 信息交互方法和装置、电子设备和存储介质
CN112818109A (zh) 邮件的智能回复方法、介质、装置和计算设备
CN108306813B (zh) 会话消息的处理方法、服务器及客户端
CN116821290A (zh) 面向多任务对话的大语言模型训练方法和交互方法
US20220129628A1 (en) Artificial intelligence system for business processes
US11361754B2 (en) Method and system for speech effectiveness evaluation and enhancement
CN110413627A (zh) 一种信息优化方法、装置、服务器及存储介质
CN112470216A (zh) 语音应用平台

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18813229

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18813229

Country of ref document: EP

Kind code of ref document: A1