Information processing method, device, electronic device and storage medium in live broadcast

Abstract

The present application provides a method, device, electronic device and computer-readable storage medium for information processing in live broadcast; it relates to natural language processing technology in the field of artificial intelligence; the method comprises: presenting the live broadcast content of the live broadcast room; in the process of presenting the live broadcast content, obtaining knowledge information from interactive events occurring in the live broadcast room; in response to a question submission operation for the live broadcast room, presenting the question submitted by the question submission operation, and obtaining the answer to the question based on the knowledge information, and presenting the answer. Through the present application, the efficiency and accuracy of answering questions in the live broadcast process can be improved.

Description

Technical Field

[0001] The present application relates to Internet technology and artificial intelligence technology, and in particular to a method, device, electronic device and computer-readable storage medium for processing information in live broadcast. Background Art

[0002] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. With the research and progress of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields.

[0003] Taking the live broadcast scenario as an example, when the audience is watching the live broadcast content they are interested in in the live broadcast room, they can ask questions to the anchor through the input box. The efficiency of replying to questions in the relevant technology is low, which affects the information dissemination effect of the live broadcast. There is no effective solution to this problem in the relevant technology. Summary of the invention

[0004] The embodiments of the present application provide a method, device, electronic device and computer-readable storage medium for information processing in a live broadcast, which can improve the efficiency and accuracy of answering questions during a live broadcast.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present application provides a method for processing information in a live broadcast, including:

[0007] Present the live broadcast content in the live broadcast room;

[0008] In the process of presenting the live broadcast content, obtaining knowledge information from interactive events occurring in the live broadcast room;

[0009] In response to a question submission operation for the live broadcast room, presenting the question submitted by the question submission operation, and

[0010] An answer to the question is obtained based on the knowledge information, and the answer is presented.

[0011] In the above scheme, after presenting the answer, the method further includes:

[0012] In response to a feedback operation on the answer, obtaining feedback information submitted by the feedback operation;

[0013] When the feedback information is of a negative type, the question is sent to a client logged in to the anchor account for presentation;

[0014] In response to an answer operation for the question, obtaining an answer submitted by the answer operation, and presenting the submitted answer;

[0015] The question and the submitted answer are combined into a question-answer pair, and the question-answer pair is synchronized to a question-answer library.

[0016] The present application provides an information processing device in a live broadcast, including:

[0017] A display module, used to present the live broadcast content of the live broadcast room;

[0018] An acquisition module, used to acquire knowledge information from interactive events occurring in the live broadcast room during the process of presenting the live broadcast content;

[0019] The answering module is used to respond to the question submission operation for the live broadcast room, present the question submitted by the question submission operation, and obtain the answer to the question based on the knowledge information and present the answer.

[0020] In the above scheme, the acquisition module is also used to acquire knowledge information in real time from interactive events occurring in the live broadcast room and / or associated live broadcast rooms; wherein, the interactive events include at least one of the following: display of recommendation information, question-and-answer events, and teaching events; wherein, the associated live broadcast rooms include at least one of the following: a live broadcast room that is in competition with the live broadcast room, a live broadcast room of an anchor that has a social relationship with the anchor in the live broadcast room, and a live broadcast room of the same type as the live broadcast room.

[0021] In the above scheme, the answer module is also used to extract multiple question-answer pairs from the knowledge information, wherein the question-answer pairs include candidate questions and corresponding candidate answers; store the multiple question-answer pairs in a question-answer library; query the question-answer library for candidate questions similar to the question, and use the candidate answers corresponding to the queried candidate questions as answers to the corresponding questions.

[0022] In the above scheme, there are multiple question and answer libraries, and the multiple question and answer libraries include a question and answer library corresponding to the live broadcast room and a question and answer library corresponding to the associated live broadcast room; the answer module is also used to query the question and answer library corresponding to the live broadcast room for candidate questions similar to the question, and use the candidate answers corresponding to the queried candidate questions as the answers to the question; when no candidate questions similar to the question are found in the question and answer library corresponding to the live broadcast room, query the question and answer library corresponding to the associated live broadcast room for candidate questions similar to the question, and use the candidate answers corresponding to the queried candidate questions as the answers to the question.

[0023] In the above scheme, the answer to the question is obtained through a machine learning model for reading comprehension; the answer module is also used to perform the following processing through the machine learning model: encoding processing is performed based on the knowledge information and the question to obtain an answer vector, and decoding processing is performed on the answer vector to obtain the answer to the question.

[0024] In the above scheme, the answer module is also used to encode the question to obtain a corresponding question vector; encode the knowledge information to obtain a corresponding knowledge vector; and perform semantic analysis on the knowledge vector based on the question vector to obtain an answer vector to the question.

[0025] In the above scheme, the answer module is also used to determine multiple key vectors and multiple value vectors corresponding to the multiple key vectors from the knowledge vector; select multiple adjacent key vectors from the multiple key vectors, and combine the multiple adjacent key vectors in multiple different ways to obtain multiple different key vector sequences; determine the attention value of each of the key vector sequences according to the question vector and the multiple value vectors; select a target key vector sequence with the largest attention value from the multiple key vector sequences, and combine the multiple key vectors contained in the target key vector sequence to obtain the answer vector.

[0026] In the above scheme, the answer module is also used to perform the following processing for each of the key vectors: determine the similarity between the question vector and the key vector, and determine the product of the similarity and the corresponding value vector as the attention value of the key vector; perform the following processing for each of the key vector sequences: add up the attention values ​​of all the key vectors contained in the key vector sequence, and determine the sum as the attention value of the key vector sequence.

[0027] In the above scheme, the answer module is also used to determine the operation of obtaining the answer to the question through the machine learning model when no candidate question suitable for the question is found in the question and answer library; wherein the question and answer library includes multiple question and answer pairs extracted from the knowledge information.

[0028] In the above scheme, the answer module is also used to retrieve the answer corresponding to the question in the question and answer library and present the retrieved answer; in response to the feedback operation on the presented answer, obtain the feedback information submitted by the feedback operation; when the feedback information is of a negative type, determine to perform an operation to obtain the answer to the question through the machine learning model; wherein the question and answer library includes multiple question and answer pairs extracted from the knowledge information.

[0029] In the above scheme, the answer module is also used to retrieve the answer to the question in the question and answer library; when the retrieved answer meets the expiration condition, it is determined to execute the operation of obtaining the answer to the question through the machine learning model; wherein the question and answer library includes multiple question and answer pairs extracted from the knowledge information; wherein the expiration condition includes at least one of the following: the length of time that the retrieved answer exists in the question and answer library exceeds a length threshold, the number of times the retrieved answer is detected does not exceed a number threshold, and the number of negative feedback information corresponding to the retrieved answer exceeds a number threshold.

[0030] In the above scheme, the information processing device in the live broadcast also includes: a synchronization module, which is used to send the question to a client logged in to the anchor account for presentation; in response to an answer operation to the question, obtain the answer submitted by the answer operation, and present the submitted answer; combine the question and the submitted answer into a question-and-answer pair, and synchronize the question-and-answer pair to a question-and-answer library.

[0031] In the above scheme, the synchronization module is also used to respond to a feedback operation on the answer, obtain feedback information submitted by the feedback operation; when the feedback information is of a negative type, send the question to a client logged in to the anchor account for presentation; respond to a solution operation on the question, obtain the answer submitted by the solution operation, and present the submitted answer; combine the question and the submitted answer into a question-and-answer pair, and synchronize the question-and-answer pair to the question-and-answer library.

[0032] An embodiment of the present application provides an electronic device, including:

[0033] A memory for storing computer executable instructions;

[0034] The processor is used to implement the information processing method in live broadcast provided by the embodiment of the present application when executing the computer executable instructions stored in the memory.

[0035] An embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions for implementing the information processing method in live broadcast provided by the embodiment of the present application when executed by a processor.

[0036] The embodiments of the present application have the following beneficial effects:

[0037] Knowledge information is learned from interactive events occurring in the live broadcast room and used to respond to questions raised in the live broadcast room. This fully utilizes the characteristic that interactive events can hit important information related to the live broadcast room, thereby improving the accuracy and efficiency of answers. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 11 is a schematic diagram of the architecture of the information processing system 100 in live broadcast provided in an embodiment of the present application;

[0039] Figure 2 is a schematic diagram of the structure of the terminal 400 provided in an embodiment of the present application;

[0040] Figure 3 It is a flowchart of the information processing method in live broadcast provided by the embodiment of the present application;

[0041] Figure 4 It is a flowchart of the information processing method in live broadcast provided by the embodiment of the present application;

[0042] Figure 5 It is a flowchart of the information processing method in live broadcast provided by the embodiment of the present application;

[0043] Figure 6 It is a schematic diagram of the principle of the information processing method in live broadcast provided by the embodiment of the present application;

[0044] Fig. 7A , Figure 7B and Figure 7C It is a schematic diagram of the principle of the information processing method in live broadcast provided by the embodiment of the present application;

[0045] Fig. 8A , Figure 8B and Figure 8C It is a schematic diagram of the principle of the information processing method in live broadcast provided by the embodiment of the present application;

[0046] Fig.9A , Fig. 9B , Fig. 9C , Fig.9D and Fig.9E It is a schematic diagram of an application scenario of the information processing method in live broadcast provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0048] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0050] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0051] 1) In response, it is used to indicate the conditions or states on which the executed operations depend. When the dependent conditions or states are met, one or more operations executed may be in real time or have a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations executed are executed.

[0052] 2) Client: an application running in the terminal to provide various services, such as a live broadcast client or a video client.

[0053] 3) Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language people use in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.

[0054] 4) Machine Learning (ML) is a multi-disciplinary interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0055] The popularity of live broadcasting has spawned many new businesses, such as live broadcasting with goods or online education. Users can ask questions to anchors or teachers through the input box, and the questions raised by users are basically answered manually, which is time-consuming and labor-intensive.

[0056] The embodiment of the present application provides a search-based question-answering system, which finds the question (Question) closest to the query (Query) in a question-answer (QA) pair (or question-answer pair) data set customized by the host or teacher, and recommends the answer corresponding to the Question to the user.

[0057] The embodiment of the present application can solve some of the problems that can be matched with the customized QA pairs. However, the coverage rate is low and it cannot solve new problems that users are interested in during the live broadcast, resulting in low efficiency in answering questions.

[0058] In response to the above technical problems, an embodiment of the present application provides an information processing method in a live broadcast, which can improve the efficiency and accuracy of answering questions during the live broadcast. The following describes an exemplary application of the information processing method in a live broadcast provided by an embodiment of the present application. The information processing method in a live broadcast provided by an embodiment of the present application can be implemented by various electronic devices. For example, it can be applied to various types of user terminals (hereinafter also referred to as terminals), such as smart phones, tablet computers, vehicle-mounted terminals, smart wearable devices, laptop computers, and desktop computers. Next, an exemplary application when the electronic device is implemented as a terminal will be described.

[0059] See also Figure 1 , Figure 1 1 is a schematic diagram of the architecture of the information processing system 100 in live broadcast provided by the embodiment of the present application. The information processing system 100 in live broadcast includes: a server 200, a network 300, and a terminal 400, which will be described separately.

[0060] The server 200 is a backend server of the client 410 , and is used to respond to the live content acquisition request of the client 410 and send the live content of the live broadcast room to the client 410 .

[0061] The network 300 is used as a medium for communication between the server 200 and the terminal 400, and may be a wide area network or a local area network, or a combination of the two.

[0062] The terminal 400 is used to run the client 410, which is a client with a live broadcast function, such as a live broadcast client or a video client. The client 410 is used to receive the live broadcast content sent by the server 200, and present the live broadcast content of the live broadcast room on the live broadcast page; it is also used to obtain knowledge information from interactive events occurring in the live broadcast room during the process of presenting the live broadcast content; it is also used to respond to the question submission operation, present the submitted question on the live broadcast page, and obtain the answer to the question based on the knowledge information and present the answer.

[0063] In some embodiments, the terminal 400 implements the information processing method in the live broadcast provided by the embodiments of the present application by running a computer program. For example, the computer program can be a native program or software module in the operating system; it can be a native application (APP, Application), that is, a program that needs to be installed in the operating system to run, such as a live broadcast APP or a video APP; it can also be a small program, that is, a program that can be run only by downloading it to a browser environment; it can also be a live broadcast small program or a video small program that can be embedded in any APP. In short, the above-mentioned computer program can be an application, module or plug-in in any form.

[0064] The embodiments of the present application can be implemented with the help of cloud technology. Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and network within a wide area network or a local area network to achieve data calculation, storage, processing, and sharing.

[0065] Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, and application technology based on the cloud computing business model. It can form a resource pool that can be used on demand and is flexible and convenient. Cloud computing technology will become an important support. The background services of the technical network system require a large amount of computing and storage resources.

[0066] As an example, the server 200 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal 400 and the server 200 may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiments of the present application.

[0067] Next, explain Figure 1 The structure of the terminal 400 in FIG. Figure 2 , Figure 2 is a schematic diagram of the structure of the terminal 400 provided in an embodiment of the present application, Figure 2 The terminal 400 shown includes: at least one processor 410, a memory 450, at least one network interface 420 and a user interface 430. The various components in the terminal 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 440 is not shown in FIG. Figure 2 Various buses are labeled as bus system 440 .

[0068] Processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.

[0069] The user interface 430 includes one or more output devices 431 that enable presentation of media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.

[0070] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.

[0071] The memory 450 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.

[0072] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.

[0073] The operating system 451 includes system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic businesses and process hardware-based tasks.

[0074] The network communication module 452 is used to reach other computing devices via one or more (wired or wireless) network interfaces 420. Exemplary network interfaces 420 include: Bluetooth, Wireless Authentication (WiFi), and Universal Serial Bus (USB).

[0075] The presentation module 453 is used to enable presentation of information (eg, a user interface for operating peripheral devices and displaying content and information) via one or more output devices 431 (eg, a display screen, a speaker, etc.) associated with the user interface 430 .

[0076] The input processing module 454 is used to detect one or more user inputs or interactions from one of the one or more input devices 432 and translate the detected inputs or interactions.

[0077] In some embodiments, the information processing device in the live broadcast provided by the embodiments of the present application can be implemented in a software manner. Figure 2 The information processing device 455 in the live broadcast stored in the memory 450 is shown, which can be software in the form of a program and a plug-in, etc., including the following software modules: a display module 4551, an acquisition module 4552 and an answer module 4553. These modules are logical, so they can be arbitrarily combined or further split according to the functions implemented. The functions of each module will be explained below.

[0078] Below, by Figure 1 The terminal 400 in the embodiment of the present application performs the information processing method in the live broadcast as an example. Figure 3 , Figure 3 is a flowchart of the information processing method in live broadcast provided by the embodiment of the present application, which will be combined with Figure 3 The steps shown are explained.

[0079] It should be noted that Figure 3 The method shown can be executed by various forms of computer programs running on the terminal 400, and is not limited to the above-mentioned client 410, such as the above-mentioned operating system 451, software modules and scripts. Therefore, the description below using the client as an example should not be regarded as a limitation on the embodiments of the present application.

[0080] In step S101, the live broadcast content of the live broadcast room is presented.

[0081] In some embodiments, the live broadcast platform includes multiple live broadcast rooms, and in response to a selection operation on the multiple live broadcast rooms, the live broadcast content of the selected live broadcast room is presented, so that the user can watch the live broadcast content that interests him or her.

[0082] Taking the application scenario of live streaming with goods as an example, the live content of the live broadcast room can be related content about items recommended by the anchor or the anchor's assistant.

[0083] Taking the application scenario of online education as an example, the live broadcast content of the live broadcast room can be related to the teacher’s lectures.

[0084] It should be noted that the anchors in the live broadcast scene may include teachers who teach through the live broadcast room, and the embodiments of the present application do not distinguish between them.

[0085] In step S102, in the process of presenting the live broadcast content, knowledge information is obtained from the interactive events occurring in the live broadcast room.

[0086] In some embodiments, the client can call the corresponding service of the terminal (e.g., knowledge information acquisition service) to complete the process of acquiring knowledge information through the terminal. The client can also call the corresponding service of the server (e.g., knowledge information acquisition service) to complete the process of acquiring knowledge information through the server.

[0087] In the following, the client calls the corresponding service of the terminal (e.g., knowledge information acquisition service) to complete the process of acquiring knowledge information through the terminal as an example. It should be noted that the client calls the corresponding service of the server (e.g., knowledge information acquisition service) to complete the process of acquiring knowledge information is similar to the following, and will not be repeated.

[0088] In some embodiments, during the presentation of live content, knowledge information is acquired in real time from interactive events occurring in the live broadcast room and / or associated live broadcast rooms.

[0089] As an example, the interactive event includes at least one of the following: display of recommendation information, question-and-answer event, and teaching event.

[0090] For example, when the interactive event is to display recommended information, the acquired knowledge information may include the parameter information of the items recommended by the anchor or the anchor's assistant in the live broadcast room (for example, the price of the item, the color of the item, the shape of the item, and the way to purchase the item, etc.), and of course, it may also be the parameter information of the items on the shelves in the background. When the interactive event is a question-and-answer event, the question-and-answer event may occur between the anchor (or teacher) and the audience (or student), or between the audience (or student) and the audience (or student), for example, the anchor answers the audience's questions. When the interactive event is a teaching event, the acquired knowledge information may include the information of the teacher explaining the course in the live broadcast room (for example, the professional terms, definitions, and principles in the course, etc.), and of course, it may also be the auxiliary teaching materials uploaded by the teacher when explaining the course. In this way, the diversity and quantity of the acquired knowledge information can be improved, thereby improving the accuracy of the answers to the subsequent questions.

[0091] As an example, the associated live broadcast room includes at least one of the following: a live broadcast room that is in competition with the live broadcast room (for example, a live broadcast room where a duel (PK, Player Killing) anchor is located or a live broadcast room that recommends the same or similar items), a live broadcast room of an anchor who has a social relationship with the anchor in the live broadcast room (for example, a live broadcast room where the anchor's friend is located), a live broadcast room of the same type as the live broadcast room (for example, a live broadcast room of the same type as the course taught by the teacher. Specifically, when the live broadcast room teaches mathematics, the associated live broadcast room can be a live broadcast room that also teaches mathematics), and a live broadcast room with the same anchor (for example, the anchor's historical live broadcast room). In this way, the channels for obtaining knowledge information can be increased, thereby improving the diversity and quantity of knowledge information obtained, and then improving the accuracy of the answers to subsequent questions.

[0092] In step S103, in response to a question submission operation for the live broadcast room, the question submitted by the question submission operation is presented.

[0093] As an example, Fig.9A In the example, the user enters a question in the input box 901 and submits it, and the submitted question 902 is presented in the live page. In this way, the user can ask questions to the anchor while watching the live broadcast.

[0094] In step S104, answers to the questions are obtained based on the knowledge information and the answers are presented.

[0095] In some embodiments, after a user submits a question and obtains an answer to the question, all answers can be presented in a unified answer area (e.g. Fig.9A As shown, the answer is presented at the top of the interactive area 903); the corresponding answer can also be presented in the adjacent position of each question (such as Fig. 9B As shown, the answer is presented adjacent to the question 903).

[0096] In the embodiment of the present application, questions in the live broadcast room are automatically replied to, which can not only save human resources but also improve the efficiency of answering questions compared to manual replies in related technologies; and the answers to questions are determined based on the knowledge information obtained from the interactive events occurring in the live broadcast room, which can improve the accuracy of the answers and thus improve the efficiency of the live broadcast.

[0097] In some embodiments, the client can call the corresponding service of the terminal (e.g., answer acquisition service) to complete the answer acquisition process through the terminal. The client can also call the corresponding service of the server (e.g., answer acquisition service) to complete the answer acquisition process through the server.

[0098] As an example, when the client calls the corresponding service of the server (e.g., answer acquisition service) to complete the process of acquiring knowledge information, the replacement steps of step S104 are: sending the question to the server; the server obtains the answer to the question based on the knowledge information and sends the answer to the client; the client presents the answer.

[0099] Below, the process of obtaining the answer by the terminal is described as an example in which the client calls the corresponding service of the terminal (e.g., the answer obtaining service). It should be noted that the process of obtaining the answer by the client calling the corresponding service of the server (e.g., the answer obtaining service) is similar to the following and will not be repeated.

[0100] In some embodiments, the answer to the question is obtained through a question and answer library. Specifically, multiple question and answer pairs are extracted from the knowledge information, wherein the question and answer pairs include candidate questions and corresponding candidate answers; the multiple question and answer pairs are stored in the question and answer library; the question and answer library is queried for candidate questions similar to the question, and the candidate answers corresponding to the queried candidate questions are used as answers to the corresponding questions.

[0101] As an example, multiple question-answer pairs are extracted from knowledge information in real time, and the multiple question-answer pairs extracted in real time are stored in a question-answer library in real time. In this way, it can be ensured that the question-answer library is updated in real time according to the live broadcast content, and the question-answer pairs updated in real time are used to answer subsequent questions raised by the audience in this live broadcast. Compared with the related art that can only use question-answer pairs to answer questions raised by the audience watching the live broadcast replay, the embodiment of the present application can ensure the timeliness of the answers in the question-answer library, thereby improving the accuracy of the answers.

[0102] As an example, the similarity between the question and each candidate question in the question-answer database is determined, and the answer to the candidate question with the greatest similarity in the question-answer database is selected as the answer to the question.

[0103] For example, Fig. 7A The FAQSet in is a pre-built question-and-answer database, Qi is a standard question in the question-and-answer database (i.e., the candidate question mentioned above), and Ai is its corresponding standard answer (i.e., the candidate answer mentioned above). new It is a question raised by the user. First find the question related to Query from FAQSet. new The closest Q i , and then Q i The corresponding answer is A i The question-and-answer database can be constructed after the live broadcast starts and before the user's questions are obtained, or before the live broadcast starts.

[0104] Since the speed of retrieving answers to similar candidate questions from the question and answer library is faster than obtaining answers through a machine learning model, the embodiments of the present application can increase the speed of obtaining answers, thereby reducing the user's waiting time.

[0105] As an example, when there are multiple question and answer libraries, and the multiple question and answer libraries include a question and answer library corresponding to a live broadcast room and a question and answer library corresponding to an associated live broadcast room, querying the question and answer library for candidate questions similar to the question, and using the candidate answers corresponding to the queried candidate questions as the answers to the corresponding questions can be: querying the question and answer library corresponding to the live broadcast room for candidate questions similar to the question, and using the candidate answers corresponding to the queried candidate questions as the answers to the question; when no candidate questions similar to the question are found in the question and answer library corresponding to the live broadcast room, querying the question and answer library corresponding to the associated live broadcast room for candidate questions similar to the question, and using the candidate answers corresponding to the queried candidate questions as the answers to the question.

[0106] For example, the question-and-answer library corresponding to the live broadcast room includes multiple question-and-answer pairs extracted from knowledge information obtained in real time from interactive events occurring in the live broadcast room; the question-and-answer library corresponding to the associated live broadcast room includes multiple question-and-answer pairs extracted from knowledge information obtained in real time from interactive events occurring in the associated live broadcast room.

[0107] The embodiment of the present application obtains the answer from a question and answer library maintained based on the knowledge information of the live broadcast room. If the answer is not obtained, the answer is obtained from a question and answer library maintained based on the knowledge information of the associated live broadcast room. This can not only improve the probability of obtaining the answer, but also improve the accuracy of the answer.

[0108] In other embodiments, the answer to the question is obtained through a machine learning model for reading comprehension. Specifically, the following processing is performed through the machine learning model: encoding processing is performed based on knowledge information and the question to obtain an answer vector, and decoding processing is performed on the answer vector to obtain the answer to the question.

[0109] As an example, the machine learning model used for reading comprehension may be a question-answering network (QANet) model. The specific implementation method of obtaining answers to questions through the QANet model will be described in detail below.

[0110] As an example, encoding based on knowledge information and questions to obtain an answer vector may include encoding the question to obtain a corresponding question vector; encoding the knowledge information to obtain a corresponding knowledge vector; and semantically analyzing the knowledge vector based on the question vector to obtain an answer vector to the question.

[0111] As a successor Fig. 8AFor example, the process of obtaining the answer to the question can be: encoding the question through the encoding layer to obtain the corresponding question vector; encoding the knowledge information through the encoding layer to obtain the corresponding knowledge vector; performing semantic analysis on the knowledge vector based on the question vector through the interaction layer to obtain the answer vector of the question; decoding the answer vector through the output layer to obtain the answer to the question.

[0112] For example, performing semantic analysis on the knowledge vector based on the question vector to obtain the answer vector of the question can be to determine multiple key vectors and multiple value vectors corresponding to the multiple key vectors from the knowledge vector; select multiple adjacent key vectors from the multiple key vectors, and combine the multiple adjacent key vectors in multiple different ways to obtain multiple different key vector sequences; determine the attention value (or attention score) of each key vector sequence based on the question vector and the multiple value vectors; select a target key vector sequence with the largest attention value from the multiple key vector sequences, and combine the multiple key vectors contained in the target key vector sequence to obtain the answer vector.

[0113] For example, based on the question vector and multiple value vectors, determining the attention value of each key vector sequence can be performing the following processing for each key vector: determining the similarity between the question vector and the key vector, and determining the product of the similarity and the corresponding value vector as the attention value of the key vector; performing the following processing for each key vector sequence: adding the attention values ​​of all key vectors contained in the key vector sequence, and determining the sum as the attention value of the key vector sequence.

[0114] For example, determine key vector A, key vector B, key vector C, and value vector A corresponding to key vector A, value vector B corresponding to key vector B, and value vector C corresponding to key vector C from the knowledge vector; determine the similarity a between the problem vector and key vector A, the similarity b between the problem vector and key vector B, and the similarity c between the problem vector and key vector C. In this way, it can be calculated that the attention value of key vector A is aA, the attention value of key vector B is bB, and the attention value of key vector C is cC. By selecting multiple adjacent key vectors from multiple key vectors for combination, we can obtain key vector sequences (key vector A, key vector B), (key vector B, key vector C) and (key vector A, key vector B, key vector C); by adding the attention values ​​of all key vectors contained in the key vector sequence, we can obtain that the attention value of the key vector sequence (key vector A, key vector B) is aA+bB; the attention value of the key vector sequence (key vector B, key vector C) is bB+cC; the attention value of the key vector sequence (key vector A, key vector B, key vector C) is aA+bB+cC; select the key vector sequence with the largest attention value, and combine multiple key vectors contained in the key vector sequence with the largest attention value to obtain the answer vector.

[0115] Since the accuracy of answers obtained through machine learning models is higher than that of retrieving answers to similar candidate questions from a question and answer library, the embodiments of the present application can improve the accuracy of answers, thereby increasing user satisfaction with the answers and further improving the efficiency of live broadcasts.

[0116] As a first example, before obtaining the answer to a question through a machine learning model, the answer to the corresponding question can also be retrieved in a question and answer library, and the retrieved answer can be presented; in response to a feedback operation on the presented answer, feedback information submitted by the feedback operation is obtained; when the feedback information is of a negative type, it is determined that an operation will be performed to obtain the answer to the question through a machine learning model; wherein the question and answer library includes multiple question and answer pairs extracted from knowledge information.

[0117] For example, the types of feedback information include positive types (eg, satisfactory or the answer has solved the submitted question) and negative types (eg, unsatisfactory or the answer has not solved the submitted question).

[0118] For example, Fig. 9C In the area displaying the answer 903, a feedback button 904 may also be displayed. The feedback button 904 may be visible and operable to all viewers in the live broadcast room, or may be visible and operable only to viewers who submitted questions corresponding to the answer 903 (e.g. Fig. 9C The viewer "Wick" who submitted the question "Which should I put in the eggs or the flour first?" is visible and actionable. When the viewer triggers the "satisfied" button, it means that the viewer is satisfied with the answer "put in the flour first", that is, there is no need to obtain the answer to the question through the machine learning model; when the viewer triggers the "unsatisfied" button, it means that the viewer is dissatisfied with the answer "put in the flour first", that is, the answer to the question needs to be obtained through the machine learning model.

[0119] In an embodiment of the present application, compared with retrieving answers to similar candidate questions from a question and answer library, obtaining answers through a machine learning model consumes more resources and is slower. Therefore, only when the user feedback is dissatisfied with the answer retrieved and displayed from the question and answer library, the answer is obtained through the machine learning model. This not only reduces resource loss but also increases the speed of obtaining answers.

[0120] As a second example, before obtaining the answer to a question through a machine learning model, it is also possible to determine that an operation of obtaining the answer to the question through a machine learning model will be performed when no candidate question matching the question is found in a question and answer library; wherein the question and answer library includes multiple question and answer pairs extracted from knowledge information.

[0121] In an embodiment of the present application, compared with retrieving answers to similar candidate questions from a question and answer library, obtaining answers through a machine learning model consumes more resources and is slower. Therefore, the answer is obtained through a machine learning model only when a candidate question matching the question is not found in the question and answer library. This not only reduces resource loss but also increases the probability of obtaining answers.

[0122] As a third example, before obtaining the answer to the question through the machine learning model, the answer to the question can also be retrieved from the question and answer library; when the retrieved answer meets the expiration condition, it is determined to perform the operation of obtaining the answer to the question through the machine learning model; wherein the question and answer library includes multiple question and answer pairs extracted from knowledge information.

[0123] Here, the failure conditions include at least one of the following: the length of time the retrieved answer exists in the question and answer database exceeds a length threshold, the number of times the retrieved answer is detected does not exceed a number threshold, and the number of negative feedback information corresponding to the retrieved answer exceeds a number threshold.

[0124] For example, the duration threshold may be a default value, or a value set by the user, client, or server, or may be determined based on the duration of existence of all candidate answers in the question-and-answer database, for example, the average duration of existence of all candidate answers is used as the duration threshold. If the duration of existence of the retrieved answer in the question-and-answer database exceeds the duration threshold, it indicates that the retrieved answer is less timely and less accurate, that is, the probability that the user is satisfied with the retrieved answer is low, so discarding the answer and obtaining the answer to the question through a machine learning model can improve the accuracy of the answer, thereby improving the user's satisfaction.

[0125] For example, the times threshold may be a default value, or a value set by a user, client, or server, or may be determined based on the number of times all candidate answers in the question-and-answer database are checked, for example, the average number of times all candidate answers are checked is used as the times threshold. If the number of times the retrieved answer is checked in the question-and-answer database does not exceed the times threshold, it indicates that the retrieved answer is rarely used and has a low credibility, so discarding the answer and obtaining the answer to the question through a machine learning model can improve the accuracy of the answer, thereby improving user satisfaction.

[0126] For example, the quantity threshold may be a default value, or a value set by a user, client, or server, or may be determined based on the number of negative feedback information corresponding to all candidate answers in the question-and-answer database, for example, the average number of negative feedback information corresponding to all candidate answers is used as the quantity threshold. If the number of negative feedback information corresponding to the retrieved answer exceeds the quantity threshold, it indicates that the probability that the user is satisfied with the retrieved answer is low, so the answer is discarded, and the answer to the question is obtained through a machine learning model, which can improve the accuracy of the answer and thus improve the user's satisfaction.

[0127] In some embodiments, when the answer to a question is not obtained, the question can be sent to a client (or anchor client) logged in to the anchor account for presentation; in response to an answer operation to the question, the answer submitted by the answer operation is obtained, and the submitted answer is presented; the question and the submitted answer are combined into a question-and-answer pair, and the question-and-answer pair is synchronized to the question-and-answer library.

[0128] As an example, Fig.9D In the example, when the answer to the submitted question 902 is not obtained, the submitted question 902 and the corresponding reply box 905 are presented in the anchor client. When the anchor enters the answer in the reply box 905 and submits it, the answer 906 submitted by the anchor is presented in the live page. The submitted question 902 and the answer 906 submitted by the anchor are combined into a question-and-answer pair and synchronized to the question-and-answer library. In this way, when the audience asks the same question again, the corresponding answer can be directly obtained from the question-and-answer library, avoiding the anchor from repeating the answer.

[0129] In some embodiments, see Figure 4 , Figure 4 is a flow chart of the information processing method in live broadcast provided by the embodiment of the present application, based on Figure 3 , step S105 and step S106 may be included after step S104.

[0130] In step S105, in response to a feedback operation on the answer, feedback information submitted by the feedback operation is obtained.

[0131] As an example, Fig.9E In the example, a feedback button 904 may also be displayed in the area displaying the answer 903, and the user may trigger a "satisfied" button and a "dissatisfied" button for the answer 903 to submit corresponding feedback information. For example, when the user triggers the "satisfied" button, the submitted feedback information is of a positive type; when the user triggers the "dissatisfied" button, the submitted feedback information is of a negative type.

[0132] In step S106, when the feedback information is of a negative type, the answer submitted by the answer operation is obtained, and the submitted answer is presented.

[0133] In some embodiments, when the feedback information is of a negative type, the question is sent to a client logged in to the anchor account for presentation; in response to an answer operation to the question, the answer submitted by the answer operation is obtained, and the submitted answer is presented; the question and the submitted answer are combined into a question-and-answer pair, and the question-and-answer pair is synchronized to the question-and-answer library.

[0134] As an example, Fig.9E In the example, when the user triggers the "unsatisfied" button for answer 903, the submitted question 902 and the corresponding reply box 905 are presented in the anchor client. When the anchor enters the answer in the reply box 905 and submits it, the answer 906 submitted by the anchor is presented in the live page. The submitted question 902 and the answer 906 submitted by the anchor are combined into a question-and-answer pair and synchronized to the question-and-answer library. In this way, when the audience asks the same question again, the corresponding answer can be directly obtained from the question-and-answer library, avoiding the anchor's repeated answer.

[0135] Below, the information processing method in live broadcast provided by the embodiment of the present application is explained by taking the application scenario of live broadcast with goods as an example.

[0136] The embodiment of the present application adds an intelligent question-and-answer assistant to the live broadcast scene to answer some common questions of users, which can help the anchor reply to users' (such as live broadcast viewers) questions and avoid the anchor's repeated replies to questions, thereby improving the efficiency of live broadcast and thus increasing economic benefits.

[0137] The embodiment of the present application realizes an intelligent question-and-answer assistant through NLP technology and dialogue technology to assist the anchor in answering some questions of the user. At the same time, the intelligent question-and-answer assistant can also realize functions such as helping the anchor to collect user questions, liven up the atmosphere (for example, when a new audience enters the live broadcast room, a welcome message "Welcome XX to the live broadcast room" is sent), and new product reminders are put on the shelves in the background. Among them, the core question-and-answer module of the intelligent question-and-answer assistant searches for answers in a variety of question-and-answer data (such as anchor-customized question-and-answer pairs and anchor live broadcast content) through a variety of question-and-answer strategies (such as retrieval-based question-and-answer and reading comprehension-based question-and-answer), which can better adapt the intelligent question-and-answer to the live broadcast scene.

[0138] In some embodiments, the intelligent question-and-answer assistant can also collect new questions from users, and the anchor will give answers, and insert the new questions and answers into the question-and-answer library in real time through voice translation technology, so that next time there is a similar question, the answer (Answer) of the QA pair can be used directly to reply, which can avoid the anchor from repeatedly answering questions. At the same time, the embodiment of the present application converts the voice content in the live broadcast content that has been played into text. For some questions that cannot be solved by the question-and-answer library, the reading comprehension technology in NLP is used to find the answer in the anchor's voice text. In this way, the embodiment of the present application can add a real-time updated question-and-answer library in the live broadcast field, as well as an intelligent question-and-answer assistant that integrates a search-based question-and-answer system and a reading comprehension-based question-and-answer system to help the anchor solve user problems and improve live broadcast efficiency.

[0139] See also Figure 5 , Figure 5 is a flowchart of the information processing method in live broadcast provided by the embodiment of the present application, which will be combined with Figure 5 Provide explanation.

[0140] In some embodiments, the user first asks a question through the input box, and then the intelligent question-and-answer assistant confirms that the input content is a question that needs to be answered, rather than a chat content, through sentence analysis. When the input content is a question that needs to be answered, the live broadcast system submits the question to the question-and-answer system to answer: First, the question-and-answer system answers through a search-based question-and-answer system, and the search-based question-and-answer system searches for the answer corresponding to the closest question in the question-and-answer library. When the search-based question-and-answer system can find the answer, it directly returns the answer to the user. When the search-based question-and-answer system cannot give an answer, it is answered by a reading comprehension-based question-and-answer system, which is mainly based on the host's explanation content and finds the answer from the explanation content. If the answer is still not found, the user's question is sent to the host, reminding the host to help answer. After the host answers, the question and answer content enters the host's explanation content again, and the question and answer pair is inserted into the question and answer library used by the search-based question-and-answer system.

[0141] The following describes a specific implementation of the information processing method in live broadcast provided by an embodiment of the present application.

[0142] (1) Sentence analysis (or question analysis).

[0143] The main purpose of sentence analysis is to determine whether the content input by the user is a question asked by the user (i.e., a question that needs to be answered) rather than a chat content. The embodiment of the present application converts the problem of sentence analysis into the problem of sentence classification, for example, the FastText model is selected for sentence binary classification.

[0144] See also Figure 6 , Figure 6is a schematic diagram of the principle of the information processing method in the live broadcast provided by the embodiment of the present application, which will be combined with Figure 6 Provide explanation.

[0145] In some embodiments, the input is a word represented by multiple vectors and its N-gram features, the output is whether the sentence is a question (for example, output 0 indicates that the sentence is a question that needs to be answered, and output 1 indicates that the sentence is not a question that needs to be answered), and the hidden layer performs superposition and average of multiple word vectors. The training data is trained by labeling a number of positive and negative samples, and a trained model is obtained through model training, and the trained model is used for model prediction.

[0146] (2) Retrieve-Based QA

[0147] The core of retrieval-based question answering is to find the most relevant or closest answer to the user's question based on the similarity between the user's question and the questions in the question and answer database.

[0148] See also Fig. 7A , Figure 7B and Figure 7C , Fig. 7A , Figure 7B and Figure 7C is a schematic diagram of the principle of the information processing method in the live broadcast provided by the embodiment of the present application, which will be combined with Fig. 7A , Figure 7B and Figure 7C Provide explanation.

[0149] In some embodiments, Fig. 7A The FAQSet in is a pre-built question-and-answer database, Qi is a standard question in the question-and-answer database (i.e., the candidate question mentioned above), and Ai is its corresponding standard answer (i.e., the candidate answer mentioned above). new is a question raised by the user. The embodiment of the present application finds the question related to Query from FAQSet. n ew The closest Q i , and finally Q i The corresponding answer is A i Return to the user.

[0150] As an example, the question-and-answer library may be constructed after the live broadcast begins and before the user's questions are obtained; or it may be constructed before the live broadcast begins.

[0151] In some embodiments, Figure 7B It is a schematic diagram of the system architecture of the retrieval-based question answering system. Figure 7BIn the retrieval-based question answering system, the system architecture is mainly divided into two parts: offline service and online service. The offline service creates an index library for questions and establishes a ranking model; after the online service obtains the user's question, it obtains multiple similar questions in the index library based on the keywords in the question, and then re-ranks multiple similar questions based on the ranking model, and selects the answer to the most similar question to display to the user.

[0152] As an example, Figure 7C This is a schematic diagram of the structure of the sorting model. The sorting model is modeled using a similarity dual-tower structure. The representation layer uses the ALBERT (A Lite BERT) model, and the matching layer uses cosine similarity. The training data uses several similar questions that are manually annotated. The training method uses loading the training model and then training through fine-tuning.

[0153] (3) Machine Reading Comprehension Question Answering (MRC-Based QA).

[0154] In some embodiments, it is assumed that the question raised by the user has appeared in the host's previous explanation, that is, the answer to the question raised by the user is limited to a clause in the article, and the model needs to mark the starting and ending positions of the correct answer in the article. Here, the definition of the QA task can be: given a text C = {c1, c2, ..., c n ,} and question Q={q1,q2,…,q m ,}, find a paragraph S from text C = {s i ,s i+1 ,…,s i+j ,} to answer question Q.

[0155] See also Fig. 8A , Figure 8B and Figure 8C , Fig. 8A , Figure 8B and Figure 8C is a schematic diagram of the principle of the information processing method in the live broadcast provided by the embodiment of the present application, which will be combined with Fig. 8A , Figure 8B and Figure 8C Provide explanation.

[0156] In some embodiments, Fig. 8A It is a system architecture diagram of the machine reading comprehension question answering system. Fig. 8AIn the example, the encoding layer encodes the questions raised by the user and the entire article. The encoding layer is used to perform underlying processing on the article and the question, respectively, and convert the text into digital codes. The interaction layer allows the model to focus on the semantic connection between the article and the question, deepen the understanding of the question through the semantic analysis of the article, and also deepen the understanding of the article through the semantic analysis of the question. The output layer generates the answer output of the model based on the semantic analysis results and the type of answer.

[0157] As an example, machine reading comprehension question answering can be implemented using the QANet model. Figure 8B The figure is a schematic diagram of the structure of the QA Net model. The QANet model consists of an input embedding layer, an embedding encoder layer, a context-query attention layer, a model encoder layer, and an output layer, which will be introduced below.

[0158] 1) Input Embedding Layer:

[0159] In some embodiments, the pre-trained word embedding is output through a highway network, and the formula is expressed as x=Highway(x w ,x c ), where x w is the word embedding vector (Word Embedding); x c It is the character embedding vector (Character Embedding), which is the output of the character-level convolutional neural network (Char-CNN).

[0160] 2) Embedding encoder layer:

[0161] In some embodiments, the embedded encoder layer includes multiple encoder blocks. Figure 8C The figure is a schematic diagram of the structure of the encoding module. The encoding module contains several convolutional neural networks (CNNs), a self-attention layer, and a feedforward layer. CNN is responsible for modeling the local text structure, and the self-attention layer is responsible for modeling the global dependency of the text. At the same time, each part adopts a residual structure.

[0162] 3) Question-Content Attention Layer:

[0163] In some embodiments, a trilinear similarity function f(q,c)=W0[q,c,q⊙c] is first defined to construct a similarity matrix S∈R between the article (Context) and the Query word n*m , and normalized by Row-wise Softmax to get the matrix Then calculate the attention matrix In addition, the method in the deep cross network (DCN) can be used to obtain the similarity matrix through Column-wise Softmax Finally, we get the question-content attention (Q2C Attention) matrix

[0164] 4) Model encoder layer:

[0165] In some embodiments, the model encoder layer takes [c, a, c⊙a, c⊙b] as input, where a and b are row vectors from attention matrices A and B, respectively. The three model encoders in the model encoder layer all adopt a bidirectional attention mechanism and share model parameters.

[0166] 5) Output layer:

[0167] In some embodiments, the starting position probability value is p 1 =softmax(W1[M0;M1]); the end position probability value is p 2 =softmax(W2[M0;M2]), where M0, M1, and M2 are the outputs of the three Model Encoders respectively.

[0168] In some embodiments, the loss function is in, is the starting position and the end position of the true answer. The meaning of this loss function is to maximize the probability of correctly predicting the beginning and end positions of the answer for each sample. As a result, the dynamic programming algorithm is used to solve the problem.

[0169] The embodiment of the present application can dynamically update the answer library. For example, the intelligent question-and-answer assistant collects questions that it cannot answer and gives them to the anchor to answer. The anchor's live broadcast content is translated into text, and the QA pairs in the live broadcast content are extracted by template matching, and the new QA pairs are inserted into the question-and-answer library.

[0170] The embodiment of the present application adds a question-and-answer system to the live broadcast scenario, which can solve users' problems conveniently and quickly, thereby improving the efficiency of the live broadcast.

[0171] Combine the following Figure 2 The exemplary structure of the information processing device in the live broadcast provided by the embodiment of the present application is described as a software module.

[0172] In some embodiments, Figure 2 As shown, the software modules stored in the information processing device 455 in the live broadcast of the memory 450 may include:

[0173] Display module 4551 is used to present the live content of the live broadcast room; acquisition module 4552 is used to acquire knowledge information from interactive events occurring in the live broadcast room during the process of presenting the live content; answer module 4553 is used to respond to the question submission operation for the live broadcast room, present the question submitted by the question submission operation, and acquire the answer to the question based on the knowledge information and present the answer.

[0174] In the above scheme, the acquisition module 4552 is also used to acquire knowledge information in real time from interactive events occurring in the live broadcast room and / or associated live broadcast rooms; wherein the interactive events include at least one of the following: display of recommended information, question-and-answer events, and teaching events; wherein the associated live broadcast room includes at least one of the following: a live broadcast room that is in competition with the live broadcast room, a live broadcast room of an anchor that has a social relationship with the anchor in the live broadcast room, and a live broadcast room of the same type as the live broadcast room.

[0175] In the above scheme, the answer module 4553 is also used to extract multiple question-answer pairs from the knowledge information, wherein the question-answer pairs include candidate questions and corresponding candidate answers; store the multiple question-answer pairs in a question-answer library; query the question-answer library for candidate questions similar to the question, and use the candidate answers corresponding to the queried candidate questions as answers to the corresponding questions.

[0176] In the above scheme, there are multiple question and answer libraries, and the multiple question and answer libraries include the question and answer library corresponding to the live broadcast room and the question and answer library corresponding to the associated live broadcast room; the answer module 4553 is also used to query the question and answer library corresponding to the live broadcast room for candidate questions similar to the question, and use the candidate answers corresponding to the queried candidate questions as the answers to the question; when no candidate questions similar to the question are found in the question and answer library corresponding to the live broadcast room, the question and answer library corresponding to the associated live broadcast room is queried for candidate questions similar to the question, and the candidate answers corresponding to the queried candidate questions are used as the answers to the question.

[0177] In the above scheme, the answer to the question is obtained through a machine learning model for reading comprehension; the answer module 4553 is also used to perform the following processing through the machine learning model: encoding processing based on knowledge information and questions to obtain an answer vector, and decoding processing on the answer vector to obtain the answer to the question.

[0178] In the above scheme, the answer module 4553 is also used to encode the question to obtain the corresponding question vector; encode the knowledge information to obtain the corresponding knowledge vector; and perform semantic analysis on the knowledge vector based on the question vector to obtain the answer vector of the question.

[0179] In the above scheme, the answer module 4553 is also used to determine multiple key vectors and multiple value vectors corresponding to the multiple key vectors from the knowledge vector; select multiple adjacent key vectors from the multiple key vectors, and combine the multiple adjacent key vectors in multiple different ways to obtain multiple different key vector sequences; determine the attention value of each key vector sequence according to the question vector and the multiple value vectors; select a target key vector sequence with the largest attention value from the multiple key vector sequences, and combine the multiple key vectors contained in the target key vector sequence to obtain an answer vector.

[0180] In the above scheme, the answer module 4553 is also used to perform the following processing for each key vector: determine the similarity between the question vector and the key vector, and determine the product of the similarity and the corresponding value vector as the attention value of the key vector; perform the following processing for each key vector sequence: add up the attention values ​​of all key vectors contained in the key vector sequence, and determine the sum as the attention value of the key vector sequence.

[0181] In the above scheme, the answer module 4553 is also used to determine the operation of obtaining the answer to the question through the machine learning model when no candidate question matching the question is found in the question and answer library; wherein the question and answer library includes multiple question and answer pairs extracted from knowledge information.

[0182] In the above scheme, the answer module 4553 is also used to retrieve the answer to the corresponding question in the question and answer library and present the retrieved answer; in response to the feedback operation on the presented answer, obtain the feedback information submitted by the feedback operation; when the feedback information is of a negative type, determine to perform an operation to obtain the answer to the question through a machine learning model; wherein the question and answer library includes multiple question and answer pairs extracted from the knowledge information.

[0183] In the above scheme, the answer module 4553 is also used to retrieve the answer to the question in the question and answer library; when the retrieved answer meets the expiration condition, it is determined to execute the operation of obtaining the answer to the question through the machine learning model; wherein the question and answer library includes multiple question and answer pairs extracted from the knowledge information; wherein the expiration condition includes at least one of the following: the length of time the retrieved answer exists in the question and answer library exceeds the length threshold, the number of times the retrieved answer is detected does not exceed the number threshold, and the number of negative feedback information corresponding to the retrieved answer exceeds the number threshold.

[0184] In the above scheme, the information processing device 455 in the live broadcast also includes: a synchronization module, which is used to send questions to the client logged in to the anchor account for presentation; in response to the answer operation to the question, obtain the answer submitted by the answer operation, and present the submitted answer; combine the question and the submitted answer into a question-answer pair, and synchronize the question-answer pair to the question-answer library.

[0185] In the above scheme, the synchronization module is also used to respond to the feedback operation on the answer, obtain the feedback information submitted by the feedback operation; when the feedback information is of a negative type, send the question to the client logged in to the anchor account for presentation; respond to the answer operation on the question, obtain the answer submitted by the answer operation, and present the submitted answer; combine the question and the submitted answer into a question-answer pair, and synchronize the question-answer pair to the question-answer library.

[0186] The embodiment of the present application provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the information processing method in the live broadcast described in the embodiment of the present application.

[0187] The embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the processor will execute the information processing method in the live broadcast provided by the embodiment of the present application, for example, Figure 3 , Figure 4 and Figure 5 The information processing method in the live broadcast is shown, and the computer includes various computing devices including intelligent terminals and servers.

[0188] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EP ROM, EEPROM, flash memory, magnetic surface storage, optical disk, or CD-ROM; or it may be various devices including one or any combination of the above memories.

[0189] In some embodiments, computer executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0190] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file storing other programs or data, such as in one or more scripts in a hypertext markup language document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0191] By way of example, computer executable instructions may be deployed to be executed on one computing device or on multiple computing devices located at one site or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0192] In summary, the embodiments of the present application have the following beneficial effects:

[0193] (1) Automatically responding to questions in the live broadcast room can not only save human resources but also improve the efficiency of answering questions compared to manual responses in related technologies.

[0194] (2) Determining the answers to questions based on the knowledge information obtained from the interactive events that occur in the live broadcast room can improve the accuracy of the answers and thus improve the efficiency of the live broadcast.

[0195] (3) Compared with obtaining answers through machine learning models, the speed of retrieving answers to similar candidate questions from the question and answer database is faster, which can increase the speed of obtaining answers and thus reduce the user's waiting time.

[0196] (4) Obtaining answers from a question-and-answer database maintained based on the knowledge information of the live broadcast room. If the answers are not obtained, then obtaining answers from a question-and-answer database maintained based on the knowledge information of the associated live broadcast room can not only improve the probability of obtaining answers, but also improve the accuracy of the answers obtained.

[0197] (5) Compared with retrieving answers to similar candidate questions from the question-and-answer database, the accuracy of answers obtained through machine learning models is higher, which can improve the accuracy of answers, thereby increasing user satisfaction with the answers and further improving the efficiency of live broadcasts.

[0198] The above is only an embodiment of the present application and is not intended to limit the protection scope of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A method for processing information in live broadcast, characterized in that: The method comprises: Present the live broadcast content in the live broadcast room; In the process of presenting the live broadcast content, acquiring knowledge information from an interactive event occurring in at least one of the live broadcast room and the associated live broadcast room; In response to a question submission operation for the live broadcast room, presenting the question submitted by the question submission operation; Extracting multiple question-answer pairs from the knowledge information, storing the multiple question-answer pairs in a question-answer database, searching the question-answer database for candidate questions similar to the question, using candidate answers corresponding to the searched candidate questions as answers corresponding to the question, and presenting the answers, wherein the question-answer pairs include candidate questions and corresponding candidate answers; When the queried answer meets the expiration condition, the answer to the question is obtained through the machine learning model, wherein the method of obtaining the answer to the question through the machine learning model includes: encoding the question to obtain a corresponding question vector, encoding the knowledge information to obtain a corresponding knowledge vector, semantically analyzing the knowledge vector based on the question vector to obtain an answer vector for the question, and decoding the answer vector to obtain the answer to the question, wherein the expiration condition includes at least one of the following: the duration of the existence of the retrieved answer in the question and answer library exceeds a duration threshold, the number of times the retrieved answer is detected does not exceed a number threshold, and the number of negative feedback information corresponding to the retrieved answer exceeds a number threshold.

2. The method according to claim 1, characterized in that The interactive event includes at least one of the following: display of recommendation information, question-and-answer event, and teaching event; The associated live broadcast rooms include at least one of the following: a live broadcast room that is in competition with the live broadcast room, a live broadcast room of a host that has a social relationship with the host in the live broadcast room, and a live broadcast room of the same type as the live broadcast room.

3. The method according to claim 1, characterized in that The number of the question and answer libraries is multiple, and the multiple question and answer libraries include a question and answer library corresponding to the live broadcast room and a question and answer library corresponding to the associated live broadcast room; The step of searching the question-answer database for candidate questions similar to the question and using the candidate answers corresponding to the searched candidate questions as the answers to the question includes: Searching for candidate questions similar to the question in the question-and-answer library corresponding to the live broadcast room, and using candidate answers corresponding to the searched candidate questions as answers to the question; When no candidate question similar to the question is found in the question and answer library corresponding to the live broadcast room, a query is made in the question and answer library corresponding to the associated live broadcast room for candidate questions similar to the question, and the candidate answers corresponding to the queried candidate questions are used as the answers to the question.

4. The method according to claim 1, characterized in that: The performing semantic analysis on the knowledge vector based on the question vector to obtain the answer vector of the question includes: Determine a plurality of key vectors and a plurality of value vectors corresponding one-to-one to the plurality of key vectors from the knowledge vector; Selecting a plurality of adjacent key vectors from the plurality of key vectors, and combining the plurality of adjacent key vectors in a plurality of different ways to obtain a plurality of different key vector sequences; Determining an attention value of each of the key vector sequences according to the question vector and the multiple value vectors; A target key vector sequence with the largest attention value is selected from the multiple key vector sequences, and multiple key vectors included in the target key vector sequence are combined to obtain the answer vector.

5. The method according to claim 4, characterized in that The step of determining the attention value of each of the key vector sequences according to the question vector and the multiple value vectors includes: For each of the key vectors, the following processing is performed: determining the similarity between the question vector and the key vector, and determining the product of the similarity and the corresponding value vector as the attention value of the key vector; The following processing is performed for each of the key vector sequences: the attention values ​​of all the key vectors contained in the key vector sequence are added up, and the sum is determined as the attention value of the key vector sequence.

6. The method according to claim 1, characterized in that Before obtaining the answer to the question through the machine learning model, the method further includes: When no candidate question matching the question is found in the question-answer database, determining to execute an operation of obtaining an answer to the question through the machine learning model; The question-answer database includes a plurality of question-answer pairs extracted from the knowledge information.

7. The method according to claim 1, characterized in that Before obtaining the answer to the question through the machine learning model, the method further includes: Retrieving the answer corresponding to the question in the question and answer database, and presenting the retrieved answer; In response to a feedback operation on a presented answer, obtaining feedback information submitted by the feedback operation; When the feedback information is of a negative type, determining to perform an operation of obtaining an answer to the question through the machine learning model; The question-answer database includes a plurality of question-answer pairs extracted from the knowledge information.

8. The method according to claim 1, characterized in that When no answer to the question is obtained, the method further comprises: Send the question to the client logged in to the anchor account for presentation; In response to an answer operation for the question, obtaining an answer submitted by the answer operation, and presenting the submitted answer; The question and the submitted answer are combined into a question-answer pair, and the question-answer pair is synchronized to a question-answer library.

9. An information processing device in live broadcast, characterized in that: include: A display module, used to present the live broadcast content of the live broadcast room; An acquisition module, used for acquiring knowledge information from an interactive event occurring in at least one of the live broadcast room and the associated live broadcast room during the process of presenting the live broadcast content; An answer module, configured to respond to a question submission operation for the live broadcast room and present the question submitted by the question submission operation; Extract multiple question-answer pairs from the knowledge information, store the multiple question-answer pairs in a question-answer library, query the question-answer library for candidate questions similar to the question, use the candidate answers corresponding to the queried candidate questions as the answers to the question, and present the answers, wherein the question-answer pair includes candidate questions and corresponding candidate answers; when the queried answers meet an expiration condition, obtain the answers to the questions through a machine learning model, wherein obtaining the answers to the questions through the machine learning model includes: encoding the questions to obtain a corresponding question vector, encoding the knowledge information to obtain a corresponding knowledge vector, semantically analyzing the knowledge vector based on the question vector to obtain an answer vector to the question, and decoding the answer vector to obtain the answer to the question, wherein the expiration condition includes at least one of the following: the duration of the existence of the retrieved answer in the question-answer library exceeds a duration threshold, the number of times the retrieved answer is detected does not exceed a number threshold, and the number of negative feedback information corresponding to the retrieved answer exceeds a number threshold.

10. An electronic device, characterized in that: include: A memory for storing computer executable instructions; The processor is used to implement the information processing method in live broadcast as described in any one of claims 1 to 8 when executing the computer executable instructions stored in the memory.

11. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and when the computer executable instructions are executed, they are used to implement the information processing method in live broadcast as described in any one of claims 1 to 8.

12. A computer program product comprising computer executable instructions or a computer program, characterized in that: When the computer executable instructions or computer program are executed by a processor, the information processing method in live broadcast described in any one of claims 1 to 8 is implemented.

Images