Data processing method, device, medium and electronic equipment for human-computer interactive dialogue
By extracting and comparing the prediction and real context characteristics in chat robot dialogue, the problem of difficult evaluation of human-computer interaction dialogue quality is solved, objective evaluation of dialogue quality is achieved, and user experience and industry development are improved.
Patent Information
- Application Number
- CN202110599432.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-05-31
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-05-31
AI Technical Summary
The prior art lacks effective methods to objectively evaluate the quality of human-computer interaction dialogue, resulting in the inability to determine whether a chatbot provides a good chat experience.
By obtaining the query text and historical dialogue data to be replied, the predicted text semantic features and real text semantic features of the current dialogue round are extracted, the dialogue quality evaluation information of human-computer interactive dialogue is generated, and the two-way self-attention mechanism and machine learning model calculation is used to calculate The differences in context characteristics are evaluated in combination with prediction and real context characteristics.
It has achieved a more objective and practical evaluation of the quality of human-computer interaction dialogue, enhanced the user experience, and provided technical personnel with a basis for optimization and improvement, and promoted the development of the human-computer interaction industry.
Smart Images

Figure CN113761156B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to a data processing method for human-computer interaction dialogue, a data processing device for human-computer interaction dialogue, a computer-readable medium, and an electronic device. Background Art
[0002] Human-computer interactive dialogue is a multi-round chat between a user and an artificial intelligence device. The content of the chat is not limited to any designated task, so there is no standard answer. The intelligent chat robot and the user can fully communicate smoothly with context, which is a reflection of the robot's high conversational ability.
[0003] However, the development of chatbots is still in the exploratory stage. It is still unclear whether the chatbots have brought a good chat experience to users. To put it more directly, there is still a lack of an evaluation method that can automatically evaluate the quality of human-computer interaction dialogues.
[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present application, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention
[0005] The purpose of the present application is to provide a data processing method for human-computer interaction dialogue, a data processing device for human-computer interaction dialogue, a computer-readable medium and an electronic device, which at least to some extent overcome the technical problems in the related art such as the inability to objectively and truly evaluate the quality of human-computer interaction dialogue.
[0006] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present application.
[0007] According to one aspect of an embodiment of the present application, a method for processing data of a human-computer interactive dialogue is provided, comprising:
[0008] Acquire a query text to be replied and historical dialogue data, wherein the historical dialogue data includes multiple rounds of dialogue data of human-computer interaction dialogues generated before the query text;
[0009] Extracting predicted text semantic features of the current conversation round according to the query text and the historical conversation data, and generating a reply text for replying to the query text according to the predicted text semantic features;
[0010] Extracting the real text semantic features of the current conversation turn according to the reply text, the query text and the historical conversation data;
[0011] The dialogue quality evaluation information of the human-computer interactive dialogue is generated according to the predicted text semantic features and the real text semantic features.
[0012] According to one aspect of an embodiment of the present application, a data processing device for a human-computer interactive dialogue is provided, the device comprising:
[0013] A data acquisition module, the data acquisition module is used to acquire the query text to be replied and the historical dialogue data, the historical dialogue data including the multi-round dialogue data of the human-computer interaction dialogue generated before the query text;
[0014] A feature extraction module, the feature extraction module is connected to the data acquisition module, and is used to extract the predicted text semantic features of the current dialogue round according to the query text and the historical dialogue data, and generate a reply text for replying to the query text according to the predicted text semantic features; the feature extraction module is also used to extract the real text semantic features of the current dialogue round according to the reply text, the query text and the historical dialogue data;
[0015] An evaluation information generation module, the evaluation information generation module is connected to the feature extraction module, and is used to generate dialogue quality evaluation information of the human-computer interaction dialogue according to the predicted text semantic features and the real text semantic features.
[0016] In one embodiment of the present application, based on the above technical solution, the feature extraction module includes:
[0017] A context feature extraction unit is configured to extract features from the query text and the historical conversation data to obtain current context features in a current conversation round;
[0018] A prediction unit, configured to predict a predicted context feature in a next dialogue turn according to the current context feature;
[0019] The semantic feature extraction unit is configured to determine the predicted text semantic feature of the current dialogue turn according to the feature difference between the predicted context feature and the current context feature.
[0020] In one embodiment of the present application, based on the above technical solution, the context feature extraction unit is further configured to perform feature extraction on the reply text, the query text and the historical conversation data to obtain the real context feature in the next conversation round;
[0021] The semantic feature extraction unit is further configured to determine the real text semantic feature of the current dialogue turn according to the feature difference between the real context feature and the current context feature.
[0022] In one embodiment of the present application, based on the above technical solution, the semantic feature extraction unit is also configured to combine the reply text, the query text and the historical conversation data to form real conversation data; input the real conversation data into the encoding model to obtain real context features; the encoding model uses a bidirectional self-attention mechanism to calculate the context-related vector representation of all context features in the real conversation data, and then takes the mean of the context-related vector representations of all context features as the real context features of the real conversation data.
[0023] In one embodiment of the present application, based on the above technical solution, the feature extraction module further includes an encoding unit, which is configured to calculate the context-related vector representation of all context features in the historical dialogue data using a bidirectional self-attention mechanism, and then take the mean of the context-related vector representations of all context features as the current context feature of the historical dialogue data;
[0024] The prediction unit is also configured to obtain all context features of the historical conversation data, and input all context features of the historical conversation data into a machine learning model for training to obtain a model for predicting context features; and input the current context features into the model for predicting context features to obtain predicted context features.
[0025] In one embodiment of the present application, based on the above technical solution, the prediction unit is also configured to obtain historical conversation data, and input all question and answer data of the historical conversation data into a machine learning model for training to obtain a model for predicting reply text; and input the query data to be replied and the semantic features of the predicted text into the model for predicting reply text to obtain the reply text.
[0026] In one embodiment of the present application, based on the above technical solution, the evaluation information generation module further includes a calculation unit, which is configured to substitute the predicted text semantic feature and the real text semantic feature into the cosine formula of the two-vector angle to obtain the vector angle between the predicted text semantic feature and the real text semantic feature;
[0027] The calculation unit is further configured to divide the minimum value of the predicted text semantic feature absolute value and the real text semantic feature absolute value by the maximum value of the predicted text semantic feature absolute value and the real text semantic feature absolute value to obtain the difference value between the predicted text semantic feature and the real text semantic feature;
[0028] The calculation unit is further configured to multiply the vector angle between the predicted text semantic feature and the real text semantic feature by a difference value to obtain dialogue quality evaluation information.
[0029] According to one aspect of an embodiment of the present application, a computer-readable medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the data processing method for human-computer interactive dialogue in the above technical solution is implemented.
[0030] According to one aspect of an embodiment of the present application, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the data processing method for human-computer interaction dialogue in the above technical solution by executing the executable instructions.
[0031] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the data processing method of the human-computer interactive dialogue in the above technical solution.
[0032] In the technical solution provided in the embodiments of the present application, the present application generates dialogue quality evaluation information of human-computer interaction dialogue by predicting text semantic features and real text semantic features. Since the predicted text semantic features and the real text semantic features correspond to changes in context, the dialogue quality evaluation information of human-computer interaction dialogue generated by the present application is combined with the specific context, which can more objectively reflect the dialogue situation of human-computer interaction. The evaluation is more practical, enhances the user experience in human-computer interaction dialogue, and is more convenient for technical personnel to optimize and improve the human-computer interaction dialogue according to the dialogue quality evaluation information of the human-computer interaction dialogue, effectively promoting the development of the human-computer interaction industry.
[0033] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] The drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application. Obviously, the drawings described below are only one embodiment of the present application, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0035] Figure 1 An exemplary system architecture block diagram applying the technical solution of the present application is schematically shown.
[0036] Figure 2 A flowchart of the steps of a data processing method for a human-computer interactive dialogue in one embodiment of the present application is schematically shown.
[0037] Figure 3 The following is a schematic flowchart of a method for extracting semantic features of predicted text in one embodiment of the present application.
[0038] Figure 4 The following is a schematic flowchart of the steps of a method for extracting predicted context features in one embodiment of the present application.
[0039] Figure 5 The following is a schematic flowchart of the steps of a method for extracting reply text in one embodiment of the present application.
[0040] Figure 6 The following is a schematic flowchart of the steps of a method for extracting semantic features of real text in one embodiment of the present application.
[0041] Figure 7 The following is a schematic flowchart of the steps of a method for extracting real context features in one embodiment of the present application.
[0042] Figure 8 The following is a schematic flowchart showing the steps of a data processing method for a human-computer interactive dialogue in a specific application embodiment of the present application.
[0043] Fig. 9 The structural block diagram of the data processing device for human-computer interactive dialogue in one embodiment of the present application is schematically shown.
[0044] Fig.10 The structure block diagram of a computer system suitable for implementing an electronic device of an embodiment of the present application is schematically shown. DETAILED DESCRIPTION
[0045] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.
[0046] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.
[0047] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.
[0048] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.
[0049] 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. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.
[0050] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0051] Among them, machine learning is a major direction of artificial intelligence software. 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 inductive learning.
[0052] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. I believe that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.
[0053] The solution provided in the embodiment of the present application involves artificial intelligence human-computer interaction dialogue technology, and the human-computer interaction dialogue uses machine learning technology in artificial intelligence software technology. Below, this application will specifically illustrate the content of the embodiment of the present application through specific human-computer interaction dialogue technology.
[0054] Human-computer interaction is a conversation between a user and an AI device. The content of the conversation is not limited to any specified task, so there is no specific standard answer. There are many types of AI devices, such as smart speakers, smart chat robots, etc. The AI device and the user can fully communicate smoothly with context, which is a reflection of the AI device's high conversation ability. Smooth communication means direct and effective interaction. The AI device will immediately respond to the questions asked by the user, and there will be no situation where the AI device is stuck and does not reply or the content of the reply is irrelevant to the question. For smooth communication, AI devices can achieve it relatively easily.
[0055] Smooth communication with context will be more difficult, because although artificial intelligence devices have a high degree of intelligence, they still cannot understand the user's complex language system. Take context for example. In different contexts, the same words may represent different meanings. Context refers to the language environment, which is the background of the conversation. There are many existing contexts, such as situational context: a factor abstracted from the actual situation that affects speech activities, including the two parties of human-computer interaction, occasions (time, place), the formality of speaking, communication media, topics or registers. Speech behavior always occurs in a certain situation, and the actual situation of speech behavior (such as relevant people, events, time, place, etc.) can also help determine the meaning expressed by the language form. There is also cultural context: it can be divided into two aspects. One is cultural customs, which refers to the life pattern that has been passed down from generation to generation and has become a habit in social life. It is a collective habit of the masses in language, behavior and psychology, which is normative and restrictive to the members of the collective. The second is social norms. It refers to the various regulations and restrictions made by a society on speech communication activities.
[0056] For example, a user says to an AI device, "Today is a cloudless day, the weather is really nice today." Here, "Today is a really nice day" is a compliment to the weather, and it also shows that the user's current mentality is positive and optimistic. The corresponding context is a context that compares the customer's mood to be better and sunny. For example, a user says to an AI device, "It has been raining since this morning, and the weather is really nice today." Here, "Today is a really nice day" is a complaint about the weather, and the corresponding context shows that the user's current mentality is negative. Therefore, if the AI device gives the same answer to the user's question without identifying the user's context, the user's human-computer interaction experience will be very poor, affecting the effective conduct of human-computer interaction. Therefore, it is difficult to objectively evaluate the quality of human-computer interaction dialogue without combining the context.
[0057] Taking smart speakers as an example, the chat experience between users and smart speakers will directly affect the overall evaluation of the product. However, the development of smart speakers is still in the exploratory stage. The product party cannot determine what kind of chat experience users think is good, let alone objectively evaluate the interactive information of smart speakers. At present, the evaluation of chat robots often relies on objective factors such as behavioral data (such as the number of conversation rounds and duration) for evaluation, or only evaluates the chat skills (such as whether the reply is coherent and timely). The evaluation dimension is too rough and lacks an overall human-computer interaction dialogue quality evaluation indicator, which makes the existing human-computer interaction dialogue evaluation very subjective and not conducive to improving the human-computer interaction dialogue. Therefore, the present application proposes a solution that can evaluate the quality of human-computer interaction dialogue. The solution of the present application will be further explained below.
[0058] Figure 1 An exemplary system architecture block diagram applying the technical solution of the present application is schematically shown.
[0059] like Figure 1 As shown, the system architecture 100 may include a terminal device 110, a network 120 and a server 130. The terminal device 110 may include various electronic devices such as smart phones, tablet computers, laptops, desktop computers, etc. The server 130 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The network 120 may be a communication medium of various connection types that can provide a communication link between the terminal device 110 and the server 130, such as a wired communication link or a wireless communication link. Among them, the artificial intelligence device may be set in the terminal device 110 or in the server 130. For example, a virtual chat software is used as an artificial intelligence device.
[0060] The user can input the query text through the terminal device 110, and the server 130 receives the query text of the terminal device 110 through the network 120. The server 130 obtains the query text to be replied and the historical dialogue data, and the historical dialogue data includes the multi-round dialogue data of the human-computer interaction dialogue generated before the query text; the historical dialogue data can be stored in the server 130. The server 130 extracts the predicted text semantic features of the current dialogue round according to the query text and the historical dialogue data, and generates a reply text for replying to the query text according to the predicted text semantic features; the server 130 then extracts the real text semantic features of the current dialogue round according to the reply text, the query text and the historical dialogue data; and generates the dialogue quality evaluation information of the human-computer interaction dialogue according to the predicted text semantic features and the real text semantic features. Therefore, the present application can extract the predicted text semantic features and the real text semantic features according to the query text and the historical dialogue data stored in the server 130 after receiving the query text of the terminal device 110 through the network 120, and then use these two semantic features to generate the dialogue quality evaluation information of the human-computer interaction dialogue.
[0061] According to the implementation needs, the system architecture in the embodiment of the present application can have any number of terminal devices, networks and servers. For example, the server 130 can be a server group composed of multiple server devices. In addition, the technical solution provided in the embodiment of the present application is mainly applied to artificial intelligence devices, and the artificial intelligence device can be embedded in the terminal device 110 or in the server 130. Therefore, the technical solution provided in the embodiment of the present application can be applied to the terminal device 110, can also be applied to the server 130, or can be implemented by the terminal device 110 and the server 130 together, and this application does not make any special restrictions on this.
[0062] The above discloses the contents of the system architecture of the technical solution corresponding to the present application, and the following specifically discloses the specific method of the technical solution corresponding to the present application.
[0063] Figure 2 The flowchart schematically shows a method for processing data of a human-computer interaction dialogue in an embodiment of the present application. The method can be applied to a server 130, or to a terminal device 110, or can be performed jointly by the terminal device 110 and the server 130. The present application discloses a method for processing data of a human-computer interaction dialogue, specifically comprising steps S210-S240.
[0064] Step S210 includes: obtaining a query text to be replied and historical dialogue data, wherein the historical dialogue data includes multiple rounds of dialogue data of human-computer interaction dialogues generated before the query text.
[0065] In specific applications, there may be multiple ways to obtain the query text to be replied. For example, when the user conducts a human-computer interaction dialogue in text, the text information input by the user may be directly received as the query text to be replied; or when the user conducts a human-computer dialogue in voice, the received voice information may be converted into text data as the query text to be replied. Or when the user conducts a human-computer interaction in image input, the image may be recognized by recognition software and converted into text data as the query text to be replied.
[0066] Furthermore, there may be multiple ways to obtain historical conversation data. Exemplarily, when the historical conversation data is stored locally on the server 130 or the terminal device 110 that is the subject of the execution of this application, the user identifier of the query text to be replied can be obtained, and then the multi-round conversation data of the human-computer interaction conversation generated by the user identifier before the query text is read locally from the server 130 or the terminal device 110 as the historical conversation data. Alternatively, exemplarily, when the historical conversation data is stored in the cloud of the network to which the server 130 or the terminal device 110 is connected for communication, the user identifier of the query text to be replied can be obtained, and then the multi-round conversation data of the human-computer interaction conversation generated by the user identifier before the query text is requested from the cloud as the historical conversation data.
[0067] In addition, for any query text to be replied, the historical dialogue data may specifically include the historical input data before the query text to be replied, and the human-computer interaction reply data corresponding to the historical input data. The number of historical input data may be set according to the specific application scenario or context, and may be one or more.
[0068] Any method of obtaining the query text to be replied and the historical conversation data can be used in the present invention, and this embodiment does not limit this.
[0069] Step S220 includes extracting semantic features of the predicted text of the current conversation turn based on the query text and the historical conversation data, and generating a reply text for replying to the query text based on the semantic features of the predicted text.
[0070] In one embodiment of the present application, a method for extracting semantic features of predicted text of a current conversation turn based on a query text and historical conversation data is specifically disclosed, including steps S310 to S320, Figure 3 The following is a schematic flowchart of a method for extracting semantic features of predicted text in one embodiment of the present application.
[0071] In step S310: feature extraction is performed on the query text and the historical conversation data to obtain the current context features in the current conversation round, and the predicted context features in the next conversation round are predicted based on the current context features.
[0072] Step S310 includes two parts. The first part, such as step S410, is to obtain the current context features. The second part, such as step S420, is to obtain the predicted context features, where the predicted context features need to be obtained based on the current context features. The extraction of the predicted context features in step S310 requires steps S410-S420. Figure 4 The following is a schematic flow chart showing the steps of a method for extracting predicted context features in one embodiment of the present application.
[0073] In one embodiment of the present application, a method for extracting features from query text and historical conversation data to obtain current context features in a current conversation turn includes:
[0074] Step S410: convert the query text and historical conversation data into feature vectors; input the historical conversation data into the encoding model in the form of feature vectors to obtain current context features.
[0075] The encoding model uses a bidirectional self-attention mechanism to calculate the context-related vector representations of all context features in the historical conversation data, and then takes the mean of the context-related vector representations of all context features as the current context feature of the historical conversation data;
[0076] The encoding model uses the Transformer model, which can include multiple stack units, each of which contains multiple forward layers and multi-head self-attention layers. The Transformer model is an Encoder-Decoder model that is completely based on the attention mechanism.
[0077] This application uses the attention mechanism of the Transformer model to judge the context of human-computer interaction dialogue. The core idea of the attention mechanism is to calculate the relationship between each word in a sentence and all the words in the sentence, and then believe that the relationship between these words reflects the relevance and importance of different words in the sentence to a certain extent. By using these relationships to adjust the importance (weight) of each word, a new expression of the entire paragraph can be obtained. This new representation not only contains the word itself, but also contains the relationship between other words and this word. This new expression can be expressed as the current context of this dialogue, so it is a more global expression compared to a simple word vector. After converting the historical dialogue data and the query text into feature vectors, the application inputs them into the encoding model, and uses the Transformer model to identify the multi-round human-computer interaction dialogue data in the historical dialogue data, and then outputs the vector corresponding to the current context feature of the historical dialogue data through the attention mechanism of the Transformer model. Therefore, the current dialogue context can be represented by the current context feature. Thereby effectively combining the context to achieve that the answers made by human-computer interaction are more in line with the needs of users.
[0078] For example, the content of the historical dialogue data is: the user asks: what is the weather like today, the AI device answers: today is sunny, the temperature is 25-30°; the user continues to ask: where is it suitable to go for fun, the AI device answers: you can go to the hill or lake next to it to enjoy the scenery; the user continues to ask: what is fun in the hill, the AI device answers: there is an observation deck on the hill, you can see the whole city. The Transformer model identifies the attention mechanism based on the above historical dialogue data, calculates the relationship between each word in a sentence and all the words in the sentence, and then believes that the relationship between these words reflects the relevance and importance of different words in the sentence to a certain extent. Finally, it is identified that the user's current context is a series of questions and answers around "play", so at this time, "play" corresponds to the current context feature. The AI device can make multi-level answers and recommendations based on the current context feature of "play" to enable users to obtain more information and improve the user experience. For example, when the AI device knows that the user is "playing" the current context feature, the AI device can also recommend tickets to various places, obtain other people's travel guides for sharing, and other relevant information about the current context feature "playing". At this time, the user can obtain more desired information, which greatly improves the user experience.
[0079] After the current context features are obtained, it is also necessary to extract the predicted context features. In one embodiment of the present application, the specific method for predicting the predicted context features in the next dialogue round according to the current context features includes:
[0080] Step S420: Obtain all contextual features of historical conversation data, and input all contextual features of historical conversation data into a machine learning model for training to obtain a model for predicting contextual features; input current contextual features into the model for predicting contextual features to obtain predicted contextual features.
[0081] The historical conversation data includes all the conversation contents of the human-computer interaction conversation between the current user and the artificial intelligence device, and the conversation contents are composed of many conversations, which may be in the same context or in different contexts. After all the context features of the current user are obtained through the historical conversation data, all the context features of the historical conversation data are obtained, and then all the context features of the historical conversation data are input into the machine learning model for training. The machine learning model can be a model built based on a convolutional neural network, a recurrent neural network, etc. Among them, the way to obtain the context features here is the same as the way to obtain the context features in step S410. Through the machine learning model, the context switching habit of the current user can be known, so that a model for predicting context features can be obtained; then the current context features are input into the model for predicting context features to obtain the predicted context features. The predicted context features are the next context features corresponding to the current context features in the model for predicting context features.
[0082] For example, through the historical conversation data, it is obtained that there are 200 conversations between the current user and the artificial intelligence device. The Transformer model is used to obtain all the context features in these 200 conversations. For example, the first 20 conversations in the historical conversation data are conversations between the current user and the artificial intelligence device to query the weather and familiarize themselves with the corresponding conversations, so the context corresponding to these 20 conversations is "weather". When the 21st to 60th conversations in the historical conversation data are conversations between the current user and the artificial intelligence device to query travel recommendations, the context corresponding to the 21st to 60th conversations is "travel places". When the 61st to 100th conversations in the historical conversation data are conversations between the current user and the artificial intelligence device to query travel ticket information and air ticket information, the context corresponding to the 61st to 100th conversations is "travel consumption amount". When the 101st to 150th dialogue of the historical dialogue data is a dialogue between the current user and the artificial intelligence device to query holidays, then the context corresponding to the 101st to 150th dialogue is "holidays"; and so on, these contexts are all obtained using the Transformer model. Therefore, when the machine learning model trains the convolutional neural network on enough historical dialogue data of the current user, it can obtain a context change of the current user, for example, the current user will change to a place to play after chatting about the weather. At this time, the current context feature can be input into the model for predicting context features to obtain the predicted context feature. For example, the current context feature is "weather", and the machine learning model finds that the context after the context of "weather" in all historical dialogue data of the current user is "place to play"; at this time, the predicted context feature can be defined as "place to play". Among them, the above text data is represented by vectors in the Transformer model and the machine learning model.
[0083] In step S320: the predicted text semantic features of the current dialogue turn are determined according to the feature difference between the predicted context features and the current context features.
[0084] The predicted text semantic feature is the feature difference between the predicted context feature and the current context feature. The feature difference can be understood as the context change between the predicted context feature and the current context feature. Through the context change between the predicted context feature and the current context feature, a simple human-computer interaction dialogue logic of the current user can be predicted. For example, when a certain user mentioned in step S420 chats, the context after the context "weather" is "place to play". The purpose of extracting the predicted text semantic feature is to obtain the user's context change through prediction to a certain extent, so as to compare the real context change corresponding to the subsequent real text semantic feature, and to conduct the final human-computer interaction dialogue quality evaluation. Since the predicted text semantic feature is the predicted context feature, which is represented by a vector, the predicted text semantic feature is the vector difference between the corresponding predicted context feature and the current context feature.
[0085] Through the above steps, the semantic features of the predicted text have been obtained. Therefore, the semantic features of the real text are needed for further comparison to obtain the final evaluation of the quality of human-computer interaction dialogue. In order to obtain the semantic features of the real text, it is necessary to first obtain the reply text that replies to the query text. The specific method of extracting the reply text that replies to the query text is as follows.
[0086] In one embodiment of the present application, a method for generating a reply text for replying a query text according to a predicted text semantic feature is specifically disclosed. The method includes steps S510 to S520. Figure 5 A flowchart of a method for extracting reply text in an embodiment of the present application is schematically shown, and the specific steps are as follows.
[0087] Step S510: Obtain historical conversation data, and input all question and answer data of the historical conversation data into a machine learning model for training to obtain a model for predicting reply text;
[0088] The historical conversation data includes all the conversation contents of the current user and the artificial intelligence device in the human-computer interaction conversation, and the conversation content is composed of many conversations. By obtaining all the conversation contents of the historical conversation data, all the conversation contents of the historical conversation data are input into the machine learning model for training. The machine learning model can be a model built based on a convolutional neural network, a recurrent neural network, etc. The machine learning model can know the conversation content between the current user and the artificial intelligence device, so as to obtain a model for predicting the reply text. Of course, a large amount of historical conversation data is required for the machine learning model. By learning a large amount of historical conversation data through the machine learning model, a model for predicting the reply text can be obtained.
[0089] Step S520: inputting the query data to be replied and the semantic features of the predicted text into a model for predicting reply text to obtain the reply text.
[0090] After the model for predicting the reply text is obtained, the query data to be replied and the semantic features of the predicted text are input into the model for predicting the reply text to obtain the reply text.
[0091] For example, the query text entered by the current user is "What fun things are there nearby?" After the machine learning model learns from the historical conversation data, the corresponding answers to "What fun things are there nearby" are mostly "You can enjoy the scenery from the hills to the west or the lakes to the east." In this case, the corresponding reply text is "You can enjoy the scenery from the hills to the west or the lakes to the east."
[0092] Through the above steps, the reply text of the query text is obtained, but the real text semantic features have not been extracted. The method for extracting the real text semantic features is as follows.
[0093] Wherein step S230 includes extracting the real text semantic features of the current conversation turn according to the reply text, the query text and the historical conversation data;
[0094] In one embodiment of the present application, a step of extracting the real text semantic features of the current conversation turn according to the reply text, the query text and the historical conversation data is specifically disclosed, including steps S610 to S620, Figure 6 A flowchart of the steps of a method for extracting semantic features of real text in an embodiment of the present application is schematically shown, which specifically includes the following steps.
[0095] Step S610: extracting features from the reply text, query text, and historical conversation data to obtain real context features in the next conversation round;
[0096] In one embodiment of the present application, a method for extracting features from reply text, query text, and historical conversation data to obtain real context features in the next conversation round is specifically disclosed, including steps S710 to S720. Figure 7 A flowchart of the steps of a method for extracting real context features in an embodiment of the present application is schematically shown, which specifically includes the following steps.
[0097] Step S710: combining the reply text, the query text and the historical conversation data to form real conversation data.
[0098] When the artificial intelligence device responds to the current user's query text and obtains the response text, the above question and answer information is combined with the historical conversation data to form the real conversation data, which is the updated conversation data. For example, there are 200 historical conversation data before, and then the current user queries through the query text and obtains the response text. If the query text and the response text correspond to 10 conversations, then the real conversation data is 210 conversation data obtained by including 200 historical conversation data and 10 new conversation information.
[0099] Step S720: Input the real conversation data into the encoding model to obtain the real context features.
[0100] Among them, the encoding model uses a bidirectional self-attention mechanism to calculate the context-related vector representations of all context features in the real conversation data, and then takes the mean of the context-related vector representations of all context features as the real context features of the real conversation data.
[0101] The encoding model and the method used here are consistent with those in step S410, and the real context features can be obtained through the encoding model. The real context features represent the context of the real conversation data formed when the reply text, query text and historical conversation data are combined. The real context features correspond to the predicted context features. The predicted context features are based on predictions, while the real context features are based on real reply texts. Therefore, both can be used to represent the context of the current user. After the real context features are obtained, the real text semantic features can be obtained by the following method.
[0102] Step S620: Determine the real text semantic features of the current dialogue turn according to the feature difference between the real context features and the current context features.
[0103] The real text semantic feature is the feature difference between the real context feature and the current context feature. The feature difference can be understood as the context change between the real context feature and the current context feature. The context change between the real context feature and the current context feature can be used to determine the context change made by the current user through the reply text. The real text semantic feature is the vector difference between the real context feature and the current context feature.
[0104] For example, referring to the example in step S420, the current user's query text is "where is fun", and the artificial intelligence device replies "there are activities in the shopping mall, which is quite fun." Then, after combining this information with the historical dialogue data, real dialogue data is formed, and real context features are obtained through real dialogue data. For example, the real context feature is "the place to play is a shopping mall"; then the real text semantic feature is the change from the current context feature "weather" to the real context feature "the place to play is a shopping mall." The predicted text semantic feature mentioned in step S420 corresponds to the change from the current context feature "weather" to the predicted context feature "the place to play is an outdoor environment." Therefore, there is a certain gap between the real text semantic features and the predicted text semantic features at this time. This situation indicates that the reply text does not match the current user's contextual habits, and the current user may be dissatisfied with the content of the reply text. Therefore, the corresponding human-computer interaction dialogue quality evaluation information will be a relatively low value.
[0105] After obtaining the real text semantic features and the predicted text semantic features, we need to use these two data to generate dialogue quality evaluation information. The specific method is as follows.
[0106] Step S240 includes generating dialogue quality evaluation information of the human-computer interactive dialogue according to the predicted text semantic features and the real text semantic features.
[0107] In one embodiment of the present application, based on the above technical solution, the dialogue quality evaluation information of the human-computer interactive dialogue is generated according to the predicted text semantic features and the real text semantic features, including:
[0108] Substitute the predicted text semantic features and the real text semantic features into the cosine formula of the angle between two vectors to obtain the vector angle between the predicted text semantic features and the real text semantic features;
[0109] The difference between the predicted text semantic feature and the real text semantic feature is obtained by dividing the minimum value of the predicted text semantic feature and the real text semantic feature by the maximum value of the predicted text semantic feature and the real text semantic feature;
[0110] The vector angle between the predicted text semantic features and the true text semantic features is multiplied by the difference value to obtain the dialogue quality evaluation information.
[0111] The specific calculation formula can be as follows:
[0112]
[0113] Among them, the predicted text semantic feature is I k '; The semantic feature of the real text is I k,; The dialogue quality evaluation information is s k .
[0114] According to the following formula, s k Perform exponential operation to obtain the dialogue quality evaluation information Flow Score, where s k The same as FlowScore, both represent conversation quality evaluation information, where Flow Score specifically calculates the conversation quality evaluation value, and the conversation quality can be evaluated according to the size of the conversation quality evaluation value.
[0115]
[0116] Among them, M represents the number of sentences in the dialogue, and the dialogue quality evaluation information is s k The range is [-1,1], (s k +1) / 2 is in the range [0,1].
[0117] The above formula can be used to directly calculate the dialogue quality evaluation information of the human-computer interaction dialogue.
[0118] The content of this application will be further explained below through a specific application example. Figure 8 The flowchart schematically shows a data processing method for human-computer interactive dialogue in a specific application embodiment of the present application, which specifically includes steps S810-S870.
[0119] Step S810: inputting historical dialogue information of the human-computer interaction dialogue between the artificial intelligence device and the user in the actual application and the query text input by the user into the preprocessing module for conversion, and converting the historical dialogue information and the query text input by the user into feature vectors;
[0120] Step S820: Transcode the historical conversation information converted into vectors and the query text input by the user through the Transformer model, and output the current context representation C of the historical conversation information k ;
[0121] Step S830: Represent the current context C k Input into the pre-trained first machine learning model to predict the predicted context representation C' k+1 ;
[0122] Step S840: Characterize C according to the current context k and predictive context representation C k+1 Computing Predicted Semantic Representations of Text I k ':I k '=C' k+1- C k ;
[0123] Step S850: semantically represent the predicted text k 'Input it into the second machine learning module together with the query text input by the user to obtain a reply text based on the query text input by the user;
[0124] Step S860: Input the reply text based on the query text input by the user into the preprocessing model and the Transformer model, and output the real context representation C k+1 ;
[0125] Step S870: Represent the real context C k+1 and the current context representation C k Do the difference and calculate the true semantic representation I k , and the predicted text semantic representation I k 'Use the formula corresponding to step S240 together to calculate the dialogue quality evaluation information Flow Score. When the value of the dialogue quality evaluation information is larger, it indicates that the quality of the human-computer interaction dialogue is higher, and the evaluation is also higher. When the value of the dialogue quality evaluation information is smaller, it indicates that the quality of the human-computer interaction dialogue is lower, and the evaluation is also worse.
[0126] The present application also discloses a data processing device 900 for human-computer interactive dialogue. Fig. 9 The structural block diagram of the data processing device for human-computer interactive dialogue in one embodiment of the present application is schematically shown. Specifically, it includes:
[0127] Data acquisition module 910, the data acquisition module 910 is used to acquire the query text to be replied and the historical dialogue data, the historical dialogue data includes the multi-round dialogue data of the human-computer interaction dialogue generated before the query text;
[0128] The feature extraction module 920 is connected to the data acquisition module 910 and is used to extract the predicted text semantic features of the current dialogue round according to the query text and the historical dialogue data, and generate a reply text for replying to the query text according to the predicted text semantic features; the feature extraction module 920 is also used to extract the real text semantic features of the current dialogue round according to the reply text, the query text and the historical dialogue data;
[0129] The evaluation information generation module 930 is connected to the feature extraction module 920 and is used to generate dialogue quality evaluation information of the human-computer interaction dialogue according to the predicted text semantic features and the real text semantic features.
[0130] Based on the above scheme, the feature extraction module 920 specifically includes: a context feature extraction unit, configured to perform feature extraction on the query text and historical dialogue data to obtain the current context features in the current dialogue round; a prediction unit, configured to predict the predicted context features in the next dialogue round based on the current context features; and a semantic feature extraction unit, configured to determine the predicted text semantic features of the current dialogue round based on the feature difference between the predicted context features and the current context features.
[0131] Based on the above scheme, the context feature extraction unit is also configured to perform feature extraction on the reply text, query text and historical conversation data to obtain the real context features in the next conversation round; the semantic feature extraction unit is also configured to determine the real text semantic features of the current conversation round based on the feature differences between the real context features and the current context features.
[0132] Based on the above scheme, the feature extraction module 920 also includes an encoding unit, which is configured to use a bidirectional self-attention mechanism to calculate the context-related vector representation of all context features in the historical conversation data, and then take the mean of the context-related vector representations of all context features as the current context feature of the historical conversation data; the prediction unit is also configured to obtain all context features of the historical conversation data, and input all context features of the historical conversation data into a machine learning model for training to obtain a model for predicting context features; the current context features are input into the model for predicting context features to obtain predicted context features.
[0133] Based on the above scheme, the prediction unit is also configured to obtain historical conversation data, and input all question and answer data of the historical conversation data into the machine learning model for training to obtain a model for predicting reply text; the query data to be replied and the semantic features of the predicted text are input into the model for predicting reply text to obtain the reply text.
[0134] Based on the above scheme, the semantic feature extraction unit is also configured to combine the reply text, query text and historical conversation data to form real conversation data; the real conversation data is input into the encoding model to obtain real context features; the encoding model uses a bidirectional self-attention mechanism to calculate the context-related vector representation of all context features in the real conversation data, and then takes the mean of the context-related vector representations of all context features as the real context features of the real conversation data.
[0135] Based on the above scheme, the evaluation information generation module 930 includes a calculation unit, which is configured to substitute the predicted text semantic features and the real text semantic features into the cosine formula of the two-vector angle to obtain the vector angle between the predicted text semantic features and the real text semantic features; the calculation unit is also configured to divide the minimum value of the absolute value of the predicted text semantic features and the absolute value of the real text semantic features by the maximum value of the absolute value of the predicted text semantic features and the absolute value of the real text semantic features to obtain the difference value between the predicted text semantic features and the real text semantic features; the calculation unit is also configured to multiply the vector angle between the predicted text semantic features and the real text semantic features by the difference value to obtain dialogue quality evaluation information.
[0136] The specific details of the object search device provided in each embodiment of the present application have been described in detail in the corresponding method embodiment and will not be repeated here.
[0137] According to one aspect of an embodiment of the present application, a computer-readable medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the data processing method for human-computer interactive dialogue in the above technical solution is implemented.
[0138] According to one aspect of an embodiment of the present application, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the data processing method for human-computer interaction dialogue in the above technical solution by executing the executable instructions.
[0139] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the data processing method of the human-computer interactive dialogue in the above technical solution.
[0140] Fig.10 The structure block diagram of a computer system for implementing an electronic device according to an embodiment of the present application is schematically shown.
[0141] It should be noted that Fig.10 The computer system 1000 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0142] like Fig.10As shown, the computer system 1000 includes a central processing unit 1001 (CPU), which can perform various appropriate actions and processes according to the program stored in the read-only memory 1002 (ROM) or the program loaded from the storage part 1008 to the random access memory 1003 (RAM). Various programs and data required for system operation are also stored in the random access memory 1003. The central processing unit 1001, the read-only memory 1002 and the random access memory 1003 are connected to each other through a bus 1004. The query / output interface 1005 (Input / Output interface, i.e., I / O interface) is also connected to the bus 1004.
[0143] The following components are connected to the query / output interface 1005: a query section 1006 including a keyboard, a mouse, etc.; an output section 1007 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1008 including a hard disk, etc.; and a communication section 1009 including a network interface card such as a LAN card, a modem, etc. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the query / output interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1010 as needed, so that a computer program read therefrom is installed into the storage section 1008 as needed.
[0144] In particular, according to an embodiment of the present application, the process described in each method flow chart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program contains a program code for executing the method shown in the flow chart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1009, and / or installed from the removable medium 1011. When the computer program is executed by the central processor 1001, various functions defined in the system of the present application are executed.
[0145] This application generates dialogue quality evaluation information of human-computer interaction dialogues by predicting text semantic features and real text semantic features. Since the predicted text semantic features and real text semantic features correspond to the changes in the context, the dialogue quality evaluation information of human-computer interaction dialogues generated by this application combines the specific context, can more objectively reflect the dialogue situation of human-computer interaction, and the evaluation is more practical, which enhances the user experience in human-computer interaction dialogues, and is more convenient for technicians to optimize and improve human-computer interaction dialogues based on the dialogue quality evaluation information of human-computer interaction dialogues. For example, artificial intelligence devices with low dialogue quality evaluation of human-computer interaction dialogues can be eliminated or re-machined to eliminate and repair those with poor quality evaluation of human-computer interaction dialogues, effectively promoting the development of the human-computer interaction industry.
[0146] It should be noted that the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by an instruction execution system, device or device or used in combination with it. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.
[0147] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0148] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.
[0149] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the implementation methods of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation methods of the present application.
[0150] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary technical means in the art that are not disclosed in the present application.
[0151] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.
Claims
1. A data processing method for human-computer interactive dialogue, It is characterized in that include: Acquire a query text to be replied and historical dialogue data, wherein the historical dialogue data includes multiple rounds of dialogue data of human-computer interaction dialogues generated before the query text; Performing feature extraction on the query text and the historical conversation data to obtain current context features in a current conversation round, and predicting predicted context features in a next conversation round based on the current context features; Determining a predicted text semantic feature of a current conversation turn according to a feature difference between the predicted context feature and the current context feature, and generating a reply text for replying to the query text according to the predicted text semantic feature; Extracting features from the reply text, the query text, and the historical conversation data to obtain real context features in the next conversation round; Determining the real text semantic feature of the current dialogue turn according to the feature difference between the real context feature and the current context feature; The dialogue quality evaluation information of the human-computer interactive dialogue is generated according to the predicted text semantic features and the real text semantic features.
2. The data processing method for human-computer interactive dialogue according to claim 1, It is characterized in that Feature extraction is performed on the reply text, the query text, and the historical conversation data to obtain real context features in the next conversation round, including: Combining the reply text, the query text and the historical conversation data to form real conversation data; The real conversation data is input into the encoding model to obtain the real context features; the encoding model uses a bidirectional self-attention mechanism to calculate the context-related vector representations of all context features in the real conversation data, and then takes the mean of the context-related vector representations of all context features as the real context features of the real conversation data.
3. The data processing method for human-computer interactive dialogue according to claim 1, It is characterized in that Extracting features from the query text and the historical conversation data to obtain current context features in a current conversation round, and predicting predicted context features in a next conversation round based on the current context features, including: Converting the query text and the historical conversation data into feature vectors; The historical conversation data is input into the encoding model in the form of a feature vector to obtain a current context feature, wherein the encoding model uses a bidirectional self-attention mechanism to calculate the context-related vector representations of all context features in the historical conversation data, and then takes the mean of the context-related vector representations of all context features as the current context feature of the historical conversation data; Obtaining all contextual features of the historical conversation data, and inputting all contextual features of the historical conversation data into a machine learning model for training to obtain a model for predicting contextual features; The current context feature is input into the model of the predicted context feature to obtain the predicted context feature.
4. The data processing method for human-computer interactive dialogue according to claim 1, It is characterized in that Generating a reply text for replying the query text according to the predicted text semantic features includes: Obtaining historical conversation data, and inputting all question and answer data of the historical conversation data into a machine learning model for training, to obtain a model for predicting reply text; The query data to be replied and the semantic features of the predicted text are input into the model of the predicted reply text to obtain the reply text.
5. The data processing method for human-computer interactive dialogue according to claim 1, It is characterized in that Generating the dialogue quality evaluation information of the human-computer interactive dialogue according to the predicted text semantic features and the real text semantic features, including: Substituting the predicted text semantic feature and the real text semantic feature into the cosine formula of the angle between two vectors to obtain the vector angle between the predicted text semantic feature and the real text semantic feature; Dividing the minimum value of the predicted text semantic feature absolute value and the real text semantic feature absolute value by the maximum value of the predicted text semantic feature absolute value and the real text semantic feature absolute value to obtain the difference value between the predicted text semantic feature and the real text semantic feature; The vector angle between the predicted text semantic feature and the real text semantic feature is multiplied by the difference value to obtain dialogue quality evaluation information.
6. A data processing device for human-computer interactive dialogue, It is characterized in that include: A data acquisition module, the data acquisition module is used to acquire the query text to be replied and the historical dialogue data, the historical dialogue data including the multi-round dialogue data of the human-computer interaction dialogue generated before the query text; A feature extraction module, the feature extraction module is connected to the data acquisition module, and is used to extract features from the query text and the historical conversation data to obtain current context features in the current conversation round, and predict predicted context features in the next conversation round based on the current context features; and used to determine the predicted text semantic features of the current dialogue round according to the feature difference between the predicted context features and the current context features, and generate a reply text for replying to the query text according to the predicted text semantic features; the feature extraction module is also used to extract features from the reply text, the query text and the historical dialogue data to obtain the real context features in the next dialogue round; and for determining the real text semantic features of the current dialogue turn according to the feature difference between the real context features and the current context features; An evaluation information generation module, the evaluation information generation module is connected to the feature extraction module, and is used to generate dialogue quality evaluation information of the human-computer interaction dialogue according to the predicted text semantic features and the real text semantic features.
7. A computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for processing data of a human-computer interactive dialogue as claimed in any one of claims 1 to 5 is implemented.
8. An electronic device, It is characterized in that include: processor; as well as A memory, configured to store executable instructions of the processor; Wherein, the processor is configured to execute the data processing method for human-computer interactive dialogue described in any one of claims 1 to 5 by executing the executable instructions.
9. A computer program product, It is characterized in that The computer program product comprises computer instructions, and the computer instructions are suitable for being loaded by a processor and executing the data processing method for human-computer interactive dialogue according to any one of claims 1 to 5.
