Interaction method and device, learning machine, electronic equipment and storage medium

Abstract

The application provides an interactive method, device, learning machine, electronic equipment and storage medium, determines and generates active interaction content based on weak knowledge points in historical interaction data; initiates interaction based on the active interaction content within a preset time after the current interaction ends. The method, device, learning machine, electronic equipment and storage medium provided by the application determine and generate active interaction content based on weak knowledge points in historical interaction data; initiate interaction based on the active interaction content within a preset time after the current interaction ends, realize the answering and guiding of weak knowledge points close to the actual learning situation of the user, especially realize the answering of knowledge points that the user may have doubts about, thereby improving the interaction efficiency of the user and improving the use experience of the user.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, and in particular to an interaction method, device, learning machine, electronic device, and storage medium. Background Technology

[0002] With the release of chatbot programs, more and more people are using chatbots to replace searching for the knowledge they want. Currently, most chatbot applications are limited to a question-and-answer model where users ask questions and the chatbot provides answers.

[0003] However, this question-and-answer model can only solve the problem raised by the user at that moment. When the user encounters a similar problem later, he or she will need to ask the chatbot again, which increases the communication time cost. At the same time, the user experience is not ideal. Summary of the Invention

[0004] This invention provides an interactive method, device, learning machine, electronic device, and storage medium to address the shortcomings of low efficiency and unsatisfactory user experience in existing human-computer interaction technologies.

[0005] This invention provides an interaction method, comprising:

[0006] Generate proactive interactive content, which is determined based on weak knowledge points in historical interaction data;

[0007] Within a preset time after the end of this interaction, an interaction is initiated proactively based on the proactive interaction content.

[0008] According to an interaction method provided by the present invention, the step of actively initiating an interaction based on the active interaction content within a preset time after the end of the current interaction includes:

[0009] If the triggering conditions are met, an interaction is initiated based on the active interaction content; the triggering conditions are determined based on the user's interaction behavior within a preset time after the end of this interaction.

[0010] According to an interaction method provided by the present invention, the generation of active interactive content includes:

[0011] Obtain the interaction data for this interaction and extract the knowledge points of this interaction from the interaction data;

[0012] Based on the knowledge points of this interaction, target knowledge points are determined from the weak knowledge points in the historical interaction data;

[0013] Based on the target knowledge points, generate proactive interactive content.

[0014] According to an interaction method provided by the present invention, determining the target knowledge point from the weak knowledge points in the historical interaction data based on the knowledge point of the current interaction includes:

[0015] Based on the interaction time of the weak knowledge points in the historical interaction data and the knowledge relevance between the weak knowledge points and the knowledge points of the current interaction, the target knowledge points are determined from the weak knowledge points in the historical interaction data.

[0016] According to an interaction method provided by the present invention, the step of actively initiating an interaction based on the active interaction content further includes:

[0017] Receive user feedback data;

[0018] The user feedback data is subjected to intent detection and / or emotion detection, and the weak knowledge points are adjusted based on the detection results.

[0019] According to an interactive method provided by the present invention, the step of determining the weak knowledge points includes:

[0020] Extract interaction knowledge points from the historical interaction data;

[0021] Extract the knowledge understanding path, including the interactive knowledge points, from the knowledge graph in the interactive domain;

[0022] Select the weak knowledge points from the knowledge points of the knowledge understanding path.

[0023] According to an interactive method provided by the present invention, the step of selecting weak knowledge points from each knowledge point in the knowledge understanding path includes:

[0024] Extract the interaction intent and / or interaction emotion associated with the interaction knowledge points from the historical interaction data;

[0025] Based on the number of times the interactive knowledge points appear in the historical interaction data and / or the interaction time, as well as the interaction intent and / or interaction emotion of the interactive knowledge points, an interaction record text is constructed.

[0026] The knowledge understanding path is textualized to obtain the knowledge path text;

[0027] Based on the weakness query model, the weak knowledge points are obtained by using the interaction record text and the knowledge path text to find the weak knowledge points.

[0028] The weakness query model is trained based on sample interaction record text, sample knowledge path text, and weakness knowledge point tags in the sample knowledge understanding path corresponding to the sample knowledge path text.

[0029] The present invention also provides an interactive device, comprising:

[0030] The generation unit generates proactive interactive content, which is determined based on weak knowledge points in historical interaction data;

[0031] The active interaction unit initiates an interaction based on the active interaction content within a preset time after the current interaction ends.

[0032] The present invention also provides a learning machine, including the interactive device described above.

[0033] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the interaction method as described above.

[0034] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the interaction method as described above.

[0035] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the interaction method as described above.

[0036] The interactive method, device, learning machine, electronic device, and storage medium provided by this invention identify and generate proactive interactive content based on weak knowledge points in historical interactive data. Within a preset time after the end of the current interaction, an interaction is proactively initiated based on the proactive interactive content, thereby providing answers and guidance for weak knowledge points that are closely related to the user's actual learning situation. In particular, it provides answers to knowledge points that the user may have potential questions about, thereby improving the user's interaction efficiency and user experience. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating the interaction method provided by the present invention;

[0039] Figure 2 This is a flowchart illustrating the method for generating proactive interactive content provided by the present invention;

[0040] Figure 3This is a flowchart illustrating the method for identifying weak knowledge points provided by the present invention;

[0041] Figure 4 This is a flowchart illustrating the method for selecting weak knowledge points provided by the present invention;

[0042] Figure 5 This is a schematic diagram of the structure of the interactive device provided by the present invention;

[0043] Figure 6 This is a schematic diagram of the learning machine provided by the present invention;

[0044] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0046] Currently, most applications of chatbots are limited to a question-and-answer model where users ask questions and the chatbot provides answers. This model can only resolve the question raised at that moment. When users encounter similar problems later, they need to seek help from the chatbot again, which increases communication time costs. At the same time, the user experience is not ideal.

[0047] In response to the above problems, Figure 1 This is a flowchart illustrating the interactive method provided by the present invention, aiming to achieve efficient interaction that aligns with the user's actual learning situation. Figure 1 As shown, the method includes:

[0048] Step 110: Generate proactive interactive content, which is determined based on weak knowledge points in historical interaction data;

[0049] Here, historical interaction data can be the interaction data between the terminal and the user recorded before the current time. Historical interaction data can include the user's questions and the terminal's responses during each round of interaction. Therefore, historical interaction data can reflect the user's weak grasp of certain knowledge points. It is understood that the interaction data during the interaction process may include at least one of voice interaction data, image interaction data, and text interaction data, and the historical interaction data obtained here is preferably in text form. Thus, the voice data during the interaction process can be transcribed into text, and the image data can be processed using OCR (Optical Character Recognition) technology to obtain the corresponding text, thereby obtaining the historical interaction data in text form.

[0050] Furthermore, "weak knowledge points" here refers to knowledge points that the user has not mastered. It's understandable that when a user asks a question about the learning machine, they will interact with the machine regarding the knowledge point they are questioning. However, users often lack sufficient understanding of other related knowledge points, which they may not mention during the interaction. Therefore, weak knowledge points can be the knowledge points the user asked about in historical interaction data, or other related knowledge points. Further, weak knowledge points can be identified by extracting interactive knowledge points from historical interaction data and matching them with the corresponding knowledge understanding paths in the knowledge graph. The knowledge point closest to the interactive knowledge point in the knowledge understanding path can be considered the weak knowledge point in that historical interaction data. Alternatively, it can be derived by combining the user's interaction intent regarding the interactive knowledge point in historical interaction data with the knowledge points that are close to the interactive knowledge point in the knowledge understanding path.

[0051] Specifically, weak knowledge points from historical interaction data can be input into a natural language model, such as an LLM (Large Language Model), to generate proactive interactive content. This proactive interactive content can include answers and / or guidance regarding the weak knowledge points.

[0052] Step 120: Within a preset time after the end of this interaction, initiate an interaction based on the active interaction content.

[0053] The preset time here can be one hour, one day, or, based on the time period recorded by the learning machine when the user last used the learning machine, the preset time can be the last time the user used the learning machine.

[0054] Specifically, within a preset timeframe after the current interaction concludes, the proactive interaction content can be formatted as interactive data in at least one of the following forms: text, image, or audio, and actively interact with the user. For example, if the current interaction of the learning machine includes a question related to an interactive knowledge point in the historical interaction data, the learning machine can, in addition to answering the question, proactively interact with the user on weak knowledge points within a preset timeframe to answer potential unmastered knowledge points or guide the user to conduct further research on the interactive knowledge point.

[0055] Understandably, when the current interaction with the learning machine does not include questions related to the knowledge points in the historical interaction data, or when the learning machine has not received user interaction for a long time, it can also proactively initiate interaction. During periods when the user frequently communicates with the learning machine, it can proactively initiate interaction with the user regarding weak knowledge points from the most recent historical interaction data. For example, it can proactively interact with the user by displaying recommended text for the weak knowledge points in a pop-up window, or by displaying recommended text for the weak knowledge points in audio format.

[0056] The method provided in this invention identifies and generates proactive interactive content based on weak knowledge points in historical interaction data. Within a preset time after the end of the current interaction, an interaction is proactively initiated based on the proactive interactive content. This achieves the answering and guidance of weak knowledge points that are close to the user's actual learning situation, especially the answering of knowledge points that the user may have questions about, thereby improving the user's interaction efficiency and user experience.

[0057] Based on any of the above embodiments, step 120 includes:

[0058] If the triggering conditions are met, an interaction is initiated based on the active interaction content; the triggering conditions are determined based on the user's interaction behavior within a preset time after the end of this interaction.

[0059] Here, user interaction can be an interaction unrelated to answering the question that the user performs within a preset time after the current interaction ends, or it can be an interaction where the user does not interact with the learning machine at all. Interactions unrelated to answering the question include, for example, the user using the learning machine to listen to music or to do extracurricular reading.

[0060] Understandably, when the triggering conditions are met, the user is in a learning idle state. Initiating interaction with the user at this time regarding weak knowledge points can greatly improve communication efficiency, thereby enhancing the user's learning efficiency. Specifically, this could be half an hour after the interaction ends; or after the learning machine has been turned on but not used for a period of time; or when the user is using the learning machine for activities unrelated to answering questions, such as listening to music or reading extracurricular books; or even when the user has not initiated any further interaction within the same day after the initial interaction. Initiating interaction through proactive content can effectively address this issue.

[0061] The method provided by this invention, under the condition of satisfying the user's interactive behavior within a preset time after the end of the current interaction, actively initiates an interaction based on the active interactive content, thereby realizing timely interaction to solve the user's weak knowledge points, improving the user's interaction efficiency, and thus helping to improve the user's learning efficiency.

[0062] Based on any of the above embodiments Figure 2 This is a flowchart illustrating the method for generating proactive interactive content provided by the present invention, as shown below. Figure 2 As shown, it includes:

[0063] Step 210: Obtain the interaction data for this interaction and extract the knowledge points of this interaction from the interaction data;

[0064] Here, "interaction data" refers to the interaction data recorded when the user interacts with the learning machine. The interaction data during the interaction process can include at least one of voice interaction data, text interaction data, and image interaction data, but the interaction data obtained here is primarily in text form. Therefore, other forms of interaction data can be transcribed into text form to obtain the text-based interaction data. Furthermore, "interaction knowledge points" here refers to the knowledge points contained in the questions posed in the interaction data, reflecting the knowledge points that the user is currently interested in exploring and understanding.

[0065] Specifically, the interaction data can be obtained through the learning machine itself or its corresponding data storage terminal. Next, key elements from the interaction data can be extracted using entity recognition, and the interaction domain to which these key elements belong can be determined. Finally, by matching each node in the knowledge graph of the corresponding interaction domain with the key elements in the interaction data, the successfully matched nodes are taken as the interaction knowledge points for this interaction.

[0066] Step 220: Based on the knowledge points of this interaction, determine the target knowledge points from the weak knowledge points in the historical interaction data;

[0067] Here, the knowledge points from this interaction reflect the user's current questions or areas of interest for exploration, while the weak knowledge points in historical interaction data reflect the knowledge points where the user needs answers or guidance for exploration, especially potential areas where the user has not yet mastered the knowledge. Therefore, based on the knowledge points from this interaction, the target knowledge points identified from the weak knowledge points in historical interaction data can serve as the knowledge points that best meet the user's current knowledge acquisition needs.

[0068] Specifically, the target knowledge point can be selected from the weak knowledge points in historical interaction data that are most relevant to the knowledge point of the current interaction. Alternatively, weak knowledge points in historical interaction data whose relevance to the knowledge point of the current interaction is less than a certain threshold can be selected as target knowledge points. Another option is to select the weak knowledge point most recent to the current interaction time.

[0069] Step 230: Generate proactive interactive content based on the target knowledge points.

[0070] Specifically, a Prompt can be constructed based on the target knowledge point. The Prompt can take the form of "The following are problems encountered by students during their learning process. Please answer these problems or guide the learning: [Target Knowledge Point]". Then, the Prompt can be input into a natural language model, such as an LLM model, to generate interactive content.

[0071] The method provided in this embodiment of the invention obtains the current interaction data and extracts the knowledge points of the current interaction from the current interaction data; based on the current interaction knowledge points, it determines the target knowledge points from the weak knowledge points in the historical interaction data; based on the target knowledge points, it generates proactive interactive content, realizing an interaction that is close to the user's current learning situation, which is more conducive to the user accepting the proactive interaction initiated by the learning machine, improving communication efficiency and user experience.

[0072] Based on any of the above embodiments, step 220 includes:

[0073] Based on the interaction time of the weak knowledge points in the historical interaction data and the knowledge relevance between the weak knowledge points and the knowledge points of the current interaction, the target knowledge points are determined from the weak knowledge points in the historical interaction data.

[0074] Here, the interaction time of weak knowledge points in historical interaction data can reflect the importance of those weak knowledge points to the user, and can be obtained by analyzing the interaction times corresponding to the historical interaction data recorded by the learning machine. It is understandable that the further the interaction time of a weak knowledge point is from the current time, the more likely the user has already mastered that weak knowledge point through other learning methods. Therefore, weak knowledge points with interaction times closer to the current time are more reliable as target knowledge points for active interaction compared to those with interaction times further away.

[0075] Furthermore, the knowledge relevance between the weak knowledge points and the knowledge points in this interaction can reflect whether there is a sequential relationship in learning progress or learning logic between the weak knowledge points and the knowledge points in this interaction, whether they belong to the same chapter of the corresponding textbook or are related knowledge points. Different weights can be pre-set, and by calculating the scores of each weak knowledge point, the weak knowledge points with the highest scores or the top few scores can be selected as target knowledge points. Alternatively, the interaction time of weak knowledge points in historical interaction data and the priority of their knowledge relevance to the knowledge points in this interaction can be pre-set. For example, among weak knowledge points whose interaction time is less than a certain time threshold, the weak knowledge points with higher knowledge relevance to the knowledge points in this interaction can be selected as target knowledge points. This embodiment of the invention does not specifically limit this approach.

[0076] The method provided in this invention determines the target knowledge point from the weak knowledge points in the historical interaction data based on the interaction time of the weak knowledge point in the historical interaction data and the knowledge relevance between the weak knowledge point and the knowledge point of the current interaction. This effectively avoids active interaction based on expired weak knowledge points, making the active interaction initiated by the learning machine more closely related to the user's current learning situation, thereby increasing the user's interest in the learning machine.

[0077] Based on any of the above embodiments, after step 120, the method further includes:

[0078] Receive user feedback data;

[0079] Here, user feedback data refers to user feedback on proactive interactions based on weak knowledge points. In other words, it reflects the feedback after interaction with the target knowledge point, indicating whether the generated text of the proactive interaction content aligns with the user's interests or learning needs. Specifically, users can provide feedback via voice interaction, which can be received using voice collection devices such as microphones. Alternatively, users can provide feedback via text interaction, which can be obtained by acquiring text interaction data after interaction based on weak knowledge points. Another approach is to receive feedback on user-initiated proactive interactions with the learning machine, such as clicking on or closing a pop-up window initiated by the learning machine.

[0080] The user feedback data is subjected to intent detection and / or emotion detection, and the weak knowledge points are adjusted based on the detection results.

[0081] Here, intent detection involves identifying user feedback data. If the identified intent matches the proactive interaction content generated based on weak knowledge points, the intent detection result is passed; otherwise, it fails. It's understood that the intent identification result reflects the missing knowledge points in the proactive interaction content generated based on weak knowledge points. Therefore, when the intent detection result is failed, the intent based on weak knowledge points can be regenerated, for example, by using other candidate weak knowledge points; when the intent detection result is passed, no adjustment to the weak knowledge points is necessary.

[0082] Furthermore, the emotion detection here can utilize deep learning-based speech emotion recognition algorithms to identify emotions in user feedback data. The emotion recognition results can be categorized, for example, into satisfaction and dissatisfaction, and this categorization result can be used as the emotion detection result. Understandably, the emotion detection result here can reflect the user's satisfaction level with the proactive interactive content generated based on weak knowledge points. When the emotion detection result is dissatisfaction, the weak knowledge points can be regenerated, for example, by using other candidate weak knowledge points; when the emotion detection result is satisfaction, no adjustment to the weak knowledge points is necessary.

[0083] The method provided in this embodiment of the invention receives user feedback data; performs intent detection and / or emotion detection on the user feedback data; and adjusts the weak knowledge points based on the detection results, thereby further realizing proactive interaction that is close to the user's weak knowledge points and improving the user experience.

[0084] Based on any of the above embodiments Figure 3This is a flowchart illustrating the method for identifying weak knowledge points provided by the present invention, as shown below. Figure 3 As shown, the steps for identifying the weak knowledge points include:

[0085] Step 310: Extract interaction knowledge points in the interaction domain from the historical interaction data;

[0086] Here, interactive knowledge points refer to the knowledge points contained in the question text in the historical interactive data, such as "the radius of the earth" or "the formation process of clouds". These can reflect knowledge points that users do not have a firm grasp of but are interested in exploring and learning about.

[0087] Specifically, historical interaction data can be obtained through the learning machine itself or its corresponding data storage terminal. Entity recognition can be performed directly on the historical interaction data to determine the interaction domain to which the identified entities belong. Entities within that interaction domain can then be directly used as interaction knowledge points within that domain. Alternatively, nodes in the knowledge graph of the interaction domain can be matched with entities in the historical interaction data, and the successfully matched nodes can be used as interaction knowledge points within that interaction domain.

[0088] The interaction domain here refers to historical interaction data. It can be the domain to which the words in the historical interaction data belong. In other words, the words here can be text content in the historical interaction data that the user does not fully understand but is interested in. The user may have questions about various aspects, so the interaction domain of the historical interaction data can also be various, such as the life domain, the learning domain, the work domain, and can be further subdivided. These will not be listed here.

[0089] Step 320: Extract the knowledge understanding path, including the interactive knowledge points, from the knowledge graph under the interactive domain;

[0090] Here, a knowledge graph within an interaction domain can be composed of knowledge points within that domain as nodes, with the connections between these knowledge points forming the edges between them. The knowledge graph reflects the relationships between these knowledge points; the closer two knowledge points are, the stronger their connection. It's understandable that the knowledge graph within each interaction domain can be a pre-built local knowledge graph or a continuously updated knowledge graph.

[0091] Furthermore, the knowledge understanding path here can be a node in a knowledge graph, where, for any given knowledge point, all connected knowledge points can serve as nodes in that path. For multiple knowledge points, it can be a path in the knowledge graph that connects them. Therefore, the knowledge points in the knowledge understanding path, including interactive knowledge points, can include the interactive knowledge point itself and other related knowledge points; that is, it may contain several interactive knowledge points and several non-interactive knowledge points.

[0092] It is understandable that if a user raises a question about an interactive knowledge point, the user may also have questions about other knowledge points related to the interactive knowledge point. Therefore, the user is more likely to have questions about other knowledge points in the knowledge understanding path, including interactive knowledge points, than about other knowledge points in the knowledge graph.

[0093] Step 330: Select the weak knowledge points from the knowledge points of the knowledge understanding path.

[0094] Specifically, the knowledge point closest to the interactive knowledge point in the knowledge understanding path can be considered the weak knowledge point for that interactive knowledge point. Alternatively, it can be determined by combining the user's interaction intent regarding the interactive knowledge point from historical interaction data with the knowledge points closest to the interactive knowledge point in the knowledge understanding path. It is understandable that multiple weak knowledge points may exist within the knowledge understanding path.

[0095] The method provided in this invention extracts interactive knowledge points in the interactive domain from the historical interactive data; extracts knowledge understanding paths including the interactive knowledge points from the knowledge graph in the interactive domain; and selects weak knowledge points from each knowledge point in the knowledge understanding path. This improves the reliability of selecting weak knowledge points, thereby making proactive interactions initiated based on weak knowledge points more closely reflect the user's actual knowledge mastery.

[0096] Based on any of the above embodiments Figure 4 This is a flowchart illustrating the method for selecting weak knowledge points provided by the present invention, as shown below. Figure 4 As shown, step 330 includes:

[0097] Step 410: Extract the interaction intent and / or interaction emotion associated with the interaction knowledge point from the historical interaction data;

[0098] Here, the interaction intent associated with the interactive knowledge point refers to the specific question the user has about that knowledge point, such as "What is the principle of GPT?". Typically, there may be questions about a single interactive knowledge point from different perspectives, and the interaction intent clarifies which specific perspective the user has a question about. It's understandable that during the interaction, there may be information unrelated to the interactive knowledge point. Therefore, specifically, by extracting question statements containing interactive knowledge points from historical interaction data and performing intent recognition on these statements, we can obtain the interaction intent associated with the interactive knowledge point. This helps to identify more accurate and relevant weak knowledge points within the knowledge comprehension path, including the interactive knowledge point.

[0099] Furthermore, the "interactive sentiment" here refers to the emotional information of users contained in historical interaction data. Especially when historical interaction data contains multiple interactive knowledge points, interactive sentiment can reflect the user's level of attention to each of these points. For example, if the historical interaction data includes the questions "How is the Pythagorean theorem proved?" and "How is the Pythagorean theorem actually used in practice?", then the interactive sentiment for the second knowledge point is more urgent than that for the first, indicating that the user's attention to the second knowledge point is higher.

[0100] It's understandable that, for users, not all rounds of historical interaction data contain weak knowledge points; similarly, for a single historical interaction data point, not all interaction knowledge points contained within that data contain weak knowledge points. Therefore, the interaction sentiment associated with interaction knowledge points in historical interaction data can reflect the user's level of attention to those knowledge points. The higher the user's level of attention to a particular interaction knowledge point, the greater the likelihood that related knowledge points are the user's weak knowledge points; conversely, the lower the user's level of attention to a particular interaction knowledge point, the less likely that related knowledge points are the user's weak knowledge points.

[0101] Specifically, the interaction emotions associated with interactive knowledge points are extracted from historical interaction data. This historical interaction data can be data whose format remains unchanged; for example, if the user interacts via voice, the historical interaction data will be in voice format; if the user interacts via text, the historical interaction data will be in text format. More specifically, for voice-based historical interaction data, deep learning-based voice emotion recognition algorithms can be used to extract the interaction emotions related to interactive knowledge points. For text-based historical interaction data, natural language processing models can be used to extract the interaction emotions related to interactive knowledge points. It is understandable that for both voice and text-based historical interaction data, the interaction emotions associated with interactive knowledge points can be extracted by inputting the historical interaction data into an LLM (Large Language Models) model.

[0102] Step 420: Based on the number of times the interactive knowledge points appear in the historical interaction data and / or the interaction time, as well as the interaction intent and / or interaction emotion of the interactive knowledge points, construct the interaction record text;

[0103] Specifically, this could involve constructing a corresponding Prompt. The Prompt could take the form of, "Regarding [interactive knowledge point], at [interaction time], the user had questions about [interactive intent] and [interactive emotion] regarding [interactive knowledge point], and the interaction regarding [interactive knowledge point] has occurred [number of times] times." For example, regarding the Pythagorean theorem, on April 28, 2023, users had significant questions about the proof process of the Pythagorean theorem, and the interaction regarding the Pythagorean theorem had occurred 3 times.

[0104] The frequency of interactive knowledge points appearing in historical interaction data can be obtained by searching through that data; the interaction time of these knowledge points in historical interaction data can be obtained from the time recorded by the terminal for that historical interaction. It's understandable that the more frequently an interactive knowledge point appears in historical data, the higher the user's interest and level of attention to that knowledge point; conversely, the fewer frequently it appears, the lower the user's interest and level of attention. Similarly, the closer the interaction time of a knowledge point in historical interaction data is to the current time, the more likely the user is to have a higher interest in that knowledge point compared to those more distant from the current time.

[0105] Step 430: Textify the knowledge understanding path to obtain the knowledge path text;

[0106] Specifically, a knowledge understanding path can be a portion of a knowledge graph, specifically a graph connected by edges between knowledge points. Therefore, the interactive knowledge points included in the knowledge understanding path can be used as topics, and the other knowledge points and their connecting edges can be converted into text, resulting in the knowledge path text. For example, when the interactive knowledge point is "Pythagorean theorem," the knowledge path text could be "the definition of the Pythagorean theorem, how to prove the Pythagorean theorem, the formula of the Pythagorean theorem, and how to prove a right triangle using the Pythagorean theorem." It's understandable that the knowledge path text, like the knowledge understanding path itself, reflects the relationship between the interactive knowledge point and other knowledge points, and the knowledge understanding path can be supplemented as needed.

[0107] Step 440: Based on the weakness query model, the interaction record text and the knowledge path text are used to find the weakness knowledge points.

[0108] The weakness query model is trained based on sample interaction record text, sample knowledge path text, and weakness knowledge point tags in the sample knowledge understanding path corresponding to the sample knowledge path text.

[0109] Specifically, a corresponding Prompt can be constructed based on the interaction record text and knowledge path text. Then, the constructed Prompt is input into the weakness query model to obtain the weak knowledge points. Here, the Prompt can be in the form of "The following is a student's learning path and the questions and situations they frequently ask during their interactions with the teacher. Please infer from this what the student's weak knowledge points are: [Knowledge path text; Interaction record text]".

[0110] The weakness query model here can be a large model with strong learning capabilities, such as a natural language processing model. It can use sample interaction records and sample knowledge path text as sample data, and the weak knowledge points in the corresponding sample knowledge understanding paths as labels. Through iterative training, the weakness query model is finally obtained. Considering the small number of weak knowledge point labels in the corresponding sample knowledge understanding paths, the training phase can be divided into unsupervised training and supervised training. In unsupervised training, iterative training is performed using a large amount of sample data to adjust the model's parameters. Then, supervised training is performed using a small number of weak knowledge points in the corresponding sample knowledge understanding paths as labels, finely iterating the model's parameters to obtain an accurate weakness query model.

[0111] Here, the sample interaction record text can be obtained entirely from actual historical interaction data, or partially from interaction records synthesized using a natural language model. The sample knowledge path text can be obtained from the sample interaction record text and the corresponding knowledge graph within the interaction domain. Furthermore, the weak knowledge point tags in the sample knowledge understanding path corresponding to the sample knowledge path text can be obtained through manual annotation.

[0112] Furthermore, after obtaining the weakness query model, the interaction record text and knowledge path text can be input into the weakness query model, and the weakness knowledge points can be output through the weakness query model.

[0113] The method provided in this invention extracts the interaction intent and / or interaction emotion associated with the interaction knowledge point from the historical interaction data; constructs interaction record text based on the frequency and / or interaction time of the interaction knowledge point in the historical interaction data, as well as the interaction intent and / or interaction emotion of the interaction knowledge point. Simultaneously, the knowledge understanding path is also textualized to obtain knowledge path text. Finally, based on a trained weakness query model, the interaction record text and the knowledge path text are applied to find weaknesses, thereby obtaining the weak knowledge points. This further improves the accuracy of identifying weak knowledge points, and thus enhances the interaction experience based on weak knowledge points.

[0114] Based on any of the above embodiments, the present invention also provides an interaction method, including:

[0115] First, generate proactive interactive content; this proactive interactive content is determined based on weak knowledge points in historical interaction data.

[0116] Specifically, historical interaction data can be the historical dialogues between each user and the learning machine, with users distinguished by their login accounts. Furthermore, considering the possibility of different users logging into the same account, the expression style of different rounds of historical dialogues under the same login account can be used to differentiate between different users under the same account.

[0117] Additionally, interaction record text can be constructed based on the frequency and / or interaction time of interactive knowledge points in historical interaction data, as well as the interaction intent and / or emotion of the interactive knowledge points. Similarly, knowledge path text is obtained by textualizing the knowledge understanding path. Then, by inputting the interaction record text and knowledge path text into a model obtained by fine-tuning the LLM model, knowledge weaknesses are obtained.

[0118] Next, if the triggering conditions are met, an interaction is initiated based on the proactive interaction content.

[0119] The triggering conditions are determined based on user interaction behavior within a preset time period after the end of this interaction.

[0120] Finally, after proactive interaction, the process also includes: receiving user feedback data; performing intent detection and / or sentiment detection on the user feedback data; and adjusting weak knowledge points based on the detection results.

[0121] Based on any of the above embodiments Figure 5 This is a schematic diagram of the structure of the interactive device provided by the present invention, as shown below. Figure 5 As shown, it includes:

[0122] The generation unit 510 generates proactive interactive content, which is determined based on weak knowledge points in historical interaction data.

[0123] The active interaction unit 520 initiates an interaction based on the active interaction content within a preset time after the end of this interaction.

[0124] The device provided in this embodiment of the invention identifies and generates proactive interactive content based on weak knowledge points in historical interaction data; within a preset time after the end of the current interaction, it proactively initiates an interaction based on the proactive interactive content, thereby providing answers and guidance for weak knowledge points that are close to the user's actual learning situation, especially for knowledge points that the user may have questions about, thus improving the user's interaction efficiency and user experience.

[0125] Based on any of the above embodiments, the active interaction unit is specifically used for:

[0126] If the triggering conditions are met, an interaction is initiated based on the active interaction content; the triggering conditions are determined based on the user's interaction behavior within a preset time after the end of this interaction.

[0127] Based on any of the above embodiments, the generating unit is specifically used for:

[0128] Obtain the interaction data for this interaction and extract the knowledge points of this interaction from the interaction data;

[0129] Based on the knowledge points of this interaction, target knowledge points are determined from the weak knowledge points in the historical interaction data;

[0130] Based on the target knowledge points, generate proactive interactive content.

[0131] Based on any of the above embodiments, the generating unit is specifically used for:

[0132] Based on the interaction time of the weak knowledge points in the historical interaction data and the knowledge relevance between the weak knowledge points and the knowledge points of the current interaction, the target knowledge points are determined from the weak knowledge points in the historical interaction data.

[0133] Based on any of the above embodiments, the active interaction unit further includes a feedback unit, which is specifically used for:

[0134] Receive user feedback data;

[0135] The user feedback data is subjected to intent detection and / or emotion detection, and the weak knowledge points are adjusted based on the detection results.

[0136] Based on any of the above embodiments, the generating unit is specifically used for:

[0137] Extract interaction knowledge points from the historical interaction data;

[0138] Extract the knowledge understanding path, including the interactive knowledge points, from the knowledge graph in the interactive domain;

[0139] Select the weak knowledge points from the knowledge points of the knowledge understanding path.

[0140] Based on any of the above embodiments, the generating unit is specifically used for:

[0141] Extract the interaction intent and / or interaction emotion associated with the interaction knowledge points from the historical interaction data;

[0142] Based on the number of times the interactive knowledge points appear in the historical interaction data and / or the interaction time, as well as the interaction intent and / or interaction emotion of the interactive knowledge points, an interaction record text is constructed.

[0143] The knowledge understanding path is textualized to obtain the knowledge path text;

[0144] Based on the weakness query model, the weak knowledge points are obtained by using the interaction record text and the knowledge path text to find the weak knowledge points.

[0145] The weakness query model is trained based on sample interaction record text, sample knowledge path text, and weakness knowledge point tags in the sample knowledge understanding path corresponding to the sample knowledge path text.

[0146] Based on any of the above embodiments Figure 6 This is a schematic diagram of the learning machine provided by the present invention, as shown below. Figure 6As shown, the learning machine includes the aforementioned interactive device. Here, 610 represents the learning machine, 510 represents the generation unit, and 520 represents the active interaction unit; 510 and 520 together constitute the aforementioned interactive device.

[0147] Specifically, when users encounter unfamiliar knowledge points while answering homework or studying outside of class, they can input their questions into the learning machine. The learning machine receives the user's questions and inputs them into the LLM (Limited Learning Model) to provide answers. Within a preset time after the interaction ends, based on the knowledge points from this interaction, the machine identifies target knowledge points from weak knowledge points in historical interaction data. It then generates proactive interaction content based on these target knowledge points and initiates proactive interaction accordingly. This proactive interaction content includes answers to questions from the current interaction data, as well as explanations or guidance on weak knowledge points related to the knowledge points from this interaction.

[0148] For example, the learning machine receives a user's question, "How to prove that a triangle is an isosceles triangle?" If the machine detects that the user's potential weakness might be "the conditions that need to be met to prove that a triangle is an isosceles triangle," then within half an hour of responding to the question, the machine can push a pop-up message stating, "To prove that a triangle is an isosceles triangle, the following two conditions need to be met: Condition 1…, Condition 2…". It's understandable that the learning machine's output of "To prove that a triangle is an isosceles triangle, the following two conditions need to be met" is the interactive script addressing the user's potential weakness, while the output of "The following steps can be used to prove that a triangle is an isosceles triangle" is the response text addressing the user's question.

[0149] The learning machine provided in this embodiment of the invention identifies and generates proactive interactive content based on weak knowledge points in historical interaction data. Within a preset time after the end of the current interaction, it proactively initiates an interaction based on the proactive interactive content, thereby providing answers and guidance for weak knowledge points that are close to the user's actual learning situation. In particular, it provides answers to knowledge points that the user may have questions about, thereby improving the user's interaction efficiency and user experience.

[0150] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7As shown, the electronic device may include a processor 710, a communications interface 720, a memory 730, and a communication bus 740. The processor 710, communications interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute an interaction method. This method includes: generating proactive interaction content, which is determined based on weak knowledge points in historical interaction data; and proactively initiating an interaction based on the proactive interaction content within a preset time after the current interaction ends.

[0151] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0152] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program, the computer program being stored on a non-transitory computer-readable storage medium, the computer program being executed by a processor, the computer being able to execute the interaction methods provided by the above methods, the method including: generating proactive interaction content, the proactive interaction content being determined based on weak knowledge points in historical interaction data; and proactively initiating an interaction based on the proactive interaction content within a preset time after the end of the current interaction.

[0153] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the interaction method provided by the above methods, the method comprising: generating proactive interaction content, the proactive interaction content being determined based on weak knowledge points in historical interaction data; and proactively initiating an interaction based on the proactive interaction content within a preset time after the end of the current interaction.

[0154] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0155] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An interaction method, characterized in that, include: Obtain the interaction data for this interaction and extract the knowledge points of this interaction from the interaction data; Based on the knowledge points of this interaction, target knowledge points are identified from the weak knowledge points in the historical interaction data; Based on the target knowledge points, generate proactive interactive content; Within a preset time after the end of this interaction, an interaction is initiated proactively based on the proactive interaction content.

2. The interaction method according to claim 1, characterized in that, The step of initiating an interaction proactively within a preset time after the end of this interaction, based on the proactive interaction content, includes: If the triggering conditions are met, an interaction is initiated based on the active interaction content; the triggering conditions are determined based on the user's interaction behavior within a preset time after the end of this interaction.

3. The interaction method according to claim 1, characterized in that, The process of identifying target knowledge points from weak knowledge points in historical interaction data based on the knowledge points of this interaction includes: Based on the interaction time of the weak knowledge points in the historical interaction data and the knowledge relevance between the weak knowledge points and the knowledge points of the current interaction, the target knowledge points are determined from the weak knowledge points in the historical interaction data.

4. The interaction method according to claim 1, characterized in that, The process of initiating an interaction based on the aforementioned proactive interaction content further includes: Receive user feedback data; The user feedback data is subjected to intent detection and / or emotion detection, and the weak knowledge points are adjusted based on the detection results.

5. The interaction method according to any one of claims 1 to 4, characterized in that, The steps for identifying the weak knowledge points include: Extract interaction knowledge points from the historical interaction data; Extract the knowledge understanding path, including the interactive knowledge points, from the knowledge graph in the interactive domain; Select the weak knowledge points from the knowledge points of the knowledge understanding path.

6. The interaction method according to claim 5, characterized in that, The step of selecting weak knowledge points from each knowledge point in the knowledge understanding path includes: Extract the interaction intent and / or interaction emotion associated with the interaction knowledge points from the historical interaction data; Based on the number of times the interactive knowledge points appear in the historical interaction data and / or the interaction time, as well as the interaction intent and / or interaction emotion of the interactive knowledge points, an interaction record text is constructed. The knowledge understanding path is textualized to obtain the knowledge path text; Based on the weakness query model, the weak knowledge points are obtained by using the interaction record text and the knowledge path text to find the weak knowledge points. The weakness query model is trained based on sample interaction record text, sample knowledge path text, and weakness knowledge point tags in the sample knowledge understanding path corresponding to the sample knowledge path text.

7. An interactive device, characterized in that, include: The generation unit acquires the interaction data for this interaction and extracts the knowledge points of this interaction from the interaction data. Based on the knowledge points of this interaction, target knowledge points are identified from the weak knowledge points in the historical interaction data; Based on the target knowledge points, generate proactive interactive content; The active interaction unit initiates an interaction based on the active interaction content within a preset time after the current interaction ends.

8. A learning machine, characterized in that, Includes the interactive device as described in claim 7.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the interaction method as described in any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the interaction method as described in any one of claims 1 to 6.

Images