Chat content monitoring method

Abstract

The invention provides a chat content monitoring method, which comprises the following steps of: extracting a fusion feature vector among chat contents of users, and comparing the fusion feature vector with a knowledge base feature vector, so as to judge whether the chat contents of the users are related to learning, and further evaluate the chat quality of the users. And when the chat quality of the user is smaller than the preset threshold value, judging that the chat content of the user is poorer in correlation with learning, determining that the user is chat at the moment, and sending early warning information, thereby realizing real-time monitoring of the learning state of the user, improving the accuracy of monitoring the learning state of the user, and being capable of reminding the user and a guardian in time. And the learning efficiency of the user is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a chat content monitoring method. Background Art

[0002] In recent years, with the continuous advancement of computer technology and educational informatization, computer and artificial intelligence technology have been gradually applied to various daily educational and teaching activities. More and more intelligent devices are gradually being applied to teaching scenarios. Students can learn through intelligent devices to provide students with more convenient learning services. However, in the process of using intelligent devices to learn, students may have ineffective learning situations such as chatting with classmates, which is not conducive to students' learning.

[0003] At present, in offline teaching scenarios, students' relevant information is usually obtained by collecting students' images or audio, and then judging whether students are learning. However, this monitoring method is not suitable for intelligent device teaching scenarios. When students use intelligent devices to chat, the monitoring equipment cannot collect students' audio information, and the image information during chatting may not be much different from the image information when students are studying, and it is impossible to accurately identify the students' learning status. Therefore, a student chat monitoring solution using intelligent teaching scenarios is needed. Summary of the invention

[0004] The present disclosure provides a chat content monitoring method.

[0005] According to a first aspect of the present disclosure, a chat content monitoring method is provided, the method comprising: in response to a chat start request and a chat termination instruction initiated by a first user to a second user, determining chat information between the first user and the second user, and a fused feature vector of one or more chat contents contained in the chat information; based on the fused feature vector, using a preset recognition model, determining a knowledge base feature vector corresponding to the fused feature vector in the preset knowledge base information, so as to determine a first similarity between the chat content and the preset knowledge base information based on the fused feature vector, the knowledge base feature vector, and a standard deviation of a preset feature space; determining chat content having a first similarity greater than a preset similarity threshold as chat content belonging to the knowledge base information, so as to determine a chat quality evaluation parameter of the first user based on a first number of chat contents belonging to the preset knowledge base information, a total number of chat contents, the first similarity, and the fused feature vector; when the chat quality evaluation parameter is less than a preset chat warning threshold, sending a warning message to a guardian terminal of the first user, so as to enable the guardian of the first user to determine the learning status of the first user.

[0006] In some embodiments of the present disclosure, determining a fused feature vector of one or more chat contents contained in a chat message includes: using a preset feature editor to extract semantic feature vectors and / or formula feature vectors in the chat content; performing weighted summation of the semantic feature vectors and / or formula feature vectors to determine the fused feature vector of the chat content.

[0007] In some embodiments of the present disclosure, the method also includes: parsing preset knowledge base information to obtain a training data set, the training data set including at least one of the following: a knowledge base text segment, formula information with a grammatical graph structure, and a chart image with coordinate annotations; randomly replacing nodes in the grammatical graph structure to obtain replacement formula information, and randomly changing the position of the coordinate annotations to obtain a disrupted chart image; using the knowledge base text segment, formula information, and chart image as positive training samples, and using the replaced formula information and the disrupted chart image as negative training samples, performing a first training on a recognition model to be trained to obtain a first training recognition model; performing synonym replacement, and / or voice conversion on the knowledge base text segment to obtain a replacement text segment; and randomly discarding nodes in the grammatical graph structure to obtain missing formula information; based on the replaced text segment, the missing formula information, and the disrupted chart image, performing a second training on the first training recognition model to obtain a preset recognition model.

[0008] In some embodiments of the present disclosure, determining the first similarity between the chat content and the preset knowledge base information based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space includes: determining the first similarity based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space in combination with the following formula 1: Formula 1, in, represents the first similarity, represents the fused feature vector, represents the knowledge base feature vector, Represents the standard deviation of the preset feature space.

[0009] In some embodiments of the present disclosure, determining the chat quality evaluation parameter of the first user based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector includes: determining the dependency parameter between the chat information and the preset knowledge base information based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity, the dependency parameter is used to indicate the degree of association between the chat information and the preset knowledge base information; determining the semantic jump distance of adjacent chat contents based on the fused feature vectors in adjacent chat contents, and determining the thinking jump parameter of the chat information based on the semantic jump distance and the preset time decay weight, the thinking jump parameter is used to indicate the logical relationship between the chat contents; determining the dynamic penalty parameter based on the first number, the total number, and the threshold number of chat contents that do not belong to the preset knowledge base content, the dynamic penalty parameter is used to indicate the degree of irrelevance of the chat information to the preset knowledge base information; performing weighted summation on the dependency parameter, the semantic jump distance, and the dynamic penalty parameter to obtain the chat quality evaluation parameter of the first user.

[0010] In some embodiments of the present disclosure, determining the dependency parameter between the chat information and the preset knowledge base information based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity includes: determining the dependency parameter based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity in combination with the following formula 2: Formula 2, Where DI represents the dependency parameter, K represents the first number of chat contents belonging to the preset knowledge base information, and N represents the total number of chat contents. Indicates the first similarity of the i-th chat content.

[0011] In some embodiments of the present disclosure, determining the semantic jump distance of adjacent chat contents based on the fused feature vectors in the adjacent chat contents, and determining the thought transition parameter of the chat information based on the semantic jump distance and the preset time decay weight includes: determining the semantic jump distance of adjacent chat contents based on the fused feature vectors in the adjacent chat contents in combination with the following formula 3: Formula 3, in, represents the fused feature vector of the jth chat content, Indicates the jth chat content Chat content with Article j-1 Based on the semantic jump distance and the preset time decay weight, combined with the following formula 4, the thought transition parameter of the chat information is determined: Formula 4, Among them, CLS represents the thinking transition parameter, Indicates the semantic jump distance in the chat message The mean of Indicates the semantic jump distance in the chat message , Indicates the preset time decay weight.

[0012] In some embodiments of the present disclosure, determining the dynamic penalty parameter based on the first number, the total number, and the threshold number of chat contents that do not belong to the preset knowledge base includes: determining the second number of chat contents that do not belong to the preset knowledge base in the chat information based on the first number and the total number; determining the dynamic penalty parameter based on the second number, the total number, and the threshold number of chat contents that do not belong to the preset knowledge base in combination with the following formula 5: Formula 5, in, represents the dynamic penalty parameter, Represents the second number, Indicates the number threshold.

[0013] In some embodiments of the present disclosure, the method also includes: determining the ratio between the learning time corresponding to the chat content belonging to the preset knowledge base information and the chat time corresponding to the chat information as an effective learning parameter; when the effective learning parameter is less than the preset learning efficiency threshold, determining the learning time as warning information.

[0014] In some embodiments of the present disclosure, the warning information also includes at least one of the following: a chat start request of the first user, a chat termination instruction of the first user, a chat start time corresponding to the chat start request, and a chat termination time corresponding to the chat termination instruction.

[0015] In summary, the present disclosure proposes a chat content monitoring method, including: in response to a chat start request and a chat termination instruction initiated by a first user to a second user, determining the chat information between the first user and the second user, and a fused feature vector of one or more chat contents contained in the chat information; based on the fused feature vector, using a preset recognition model, determining a knowledge base feature vector corresponding to the fused feature vector in the preset knowledge base information, so as to determine a first similarity between the chat content and the preset knowledge base information based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space; determining the chat content whose first similarity is greater than a preset similarity threshold as the chat content belonging to the knowledge base information, so as to determine the chat quality evaluation parameter of the first user based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector; when the chat quality evaluation parameter is less than the preset chat warning threshold, sending a warning message to the guardian terminal of the first user, so that the guardian of the first user can determine the learning status of the first user. The method disclosed in the present invention extracts the fusion feature vector between the user's chat content, and compares the fusion feature vector with the knowledge base feature vector, so as to judge whether the user's chat content is related to learning, and then evaluate the user's chat quality. When the user's chat quality is less than a preset threshold, it is judged that the user's chat content is poorly related to learning, and it is determined that the user is chatting at this time, and an early warning message is issued, thereby realizing real-time monitoring of the user's learning status, improving the accuracy of monitoring the user's learning status, and being able to timely remind the user and the guardian, thereby improving the user's learning efficiency.

[0016] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure. Figure 1 A flow chart of a chat content monitoring method provided by an embodiment of the present disclosure; Figure 2 A flowchart of another chat content monitoring method provided by an embodiment of the present disclosure; Figure 3 A flowchart of another chat content monitoring method provided by an embodiment of the present disclosure; Figure 4 A flowchart of another chat content monitoring method provided by an embodiment of the present disclosure; Figure 5 A flowchart of another chat content monitoring method provided by an embodiment of the present disclosure; Figure 6 A schematic diagram of the structure of a chat content monitoring device provided by an embodiment of the present disclosure; Figure 7 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0019] In recent years, with the continuous advancement of computer technology and educational informatization, computer and artificial intelligence technology have been gradually applied to various daily educational and teaching activities. More and more intelligent devices are gradually being applied to teaching scenarios. Students can learn through intelligent devices to provide students with more convenient learning services. However, in the process of using intelligent devices to learn, students may have ineffective learning situations such as chatting with classmates, which is not conducive to students' learning.

[0020] At present, in offline teaching scenarios, students' relevant information is usually obtained by collecting students' images or audio, and then judging whether students are learning. However, this monitoring method is not suitable for intelligent device teaching scenarios. When students use intelligent devices to chat, the monitoring equipment cannot collect students' audio information, and the image information during chatting may not be much different from the image information when students are studying, and it is impossible to accurately identify the students' learning status. Therefore, a student chat monitoring solution using intelligent teaching scenarios is needed.

[0021] In order to solve the problems in the related technology, the present invention proposes a chat content monitoring method, which extracts the fusion feature vector between the user's chat content and compares the fusion feature vector with the knowledge base feature vector to determine whether the user's chat content is related to learning, and then evaluates the user's chat quality. When the user's chat quality is less than a preset threshold, it is judged that the user's chat content is poorly relevant to learning, and it is determined that the user is chatting at this time, and an early warning message is issued, thereby realizing real-time monitoring of the user's learning status, improving the accuracy of monitoring the user's learning status, and being able to promptly remind users and guardians, thereby improving the user's learning efficiency.

[0022] A chat content monitoring method according to an embodiment of the present disclosure is described below with reference to the accompanying drawings.

[0023] Figure 1 The following is a flow chart of a chat content monitoring method provided by an embodiment of the present disclosure. Figure 1 As shown, the method includes: Step 101: In response to a chat start request and a chat termination instruction initiated by a first user to a second user, determine chat information between the first user and the second user, and a fusion feature vector of one or more chat contents contained in the chat information.

[0024] In some embodiments, the first user may initiate a chat request to the second user through a smart terminal, a smart teaching system or other device, but is not limited thereto. The present disclosure does not limit the manner in which the chat is implemented between the first user and the second user.

[0025] In some embodiments, the first user is the user who initiates the chat request, and the second user is the user who accepts the chat request of the first user. In other words, the second user is the user who chats with the first user.

[0026] In some embodiments, the chat start request can be a learning assistance request initiated by the first user to the second user, that is, the first user sends a chat message to the second user to request the second user to help solve the learning problem; it can also be a learning material sharing request initiated by the first user to the second user, that is, the first user sends a message to the second user to request the second user to share learning materials with the first user, but is not limited to this. The present disclosure does not limit the specific method of requesting learning assistance.

[0027] In some embodiments, the chat termination instruction can be an instruction triggered when the first user or the second user clicks a button to exit the chat interface, or it can be an instruction triggered when the first user or the second user sends a message indicating that the problem has been solved / learning materials have been sent, etc., but is not limited to this. The present disclosure does not limit the specific form of the chat termination instruction.

[0028] In some embodiments, the chat content included between the chat start request and the chat termination instruction can be determined as chat information between the first user and the second user. In other words, the chat information is all chat messages between the first user and the second user in a complete chat start and end cycle.

[0029] In some embodiments, the one or more chat contents included in the chat information, that is, there are multiple chat messages between the first user and the second user between the chat start request and the chat termination instruction, and one of the chat contents is a chat message between the first user and the second user.

[0030] In some embodiments, a preset feature editor can be used to extract semantic feature vectors and / or formula feature vectors in the chat content, and then the semantic feature vectors and / or formula feature vectors are fused to determine a fused feature vector. For details, see Figure 2 The embodiments shown will not be described in detail here.

[0031] In some embodiments, the fused feature vector may be used to indicate the information contained in each chat content, such as the contained knowledge points, formulas, etc.

[0032] Step 102: Based on the fused feature vector, using a preset recognition model, determine the knowledge base feature vector in the preset knowledge base information that corresponds to the fused feature vector, so as to determine the first similarity between the chat content and the preset knowledge base information based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space.

[0033] In some embodiments, the preset knowledge base information may include teaching material text, formula information, chart information, etc. The present disclosure is not limited to this. It should be understood that the preset knowledge base information may be added or deleted as the teaching material and other samples change.

[0034] In some embodiments, the preset recognition model can extract relevant features from the preset knowledge base information to generate multiple knowledge base feature vectors, and then determine the knowledge base feature vector containing the same or similar content as the fused feature vector by comparing the fused feature vector with the multiple knowledge base feature vectors. A training method for the preset recognition model can be found in Figure 3 The embodiments shown will not be described in detail here.

[0035] In some embodiments, the first similarity between the fused feature vector and the corresponding knowledge base feature vector can be determined by fusing the feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space using Formula 1 to determine the similarity between the chat content and the knowledge base information: Formula 1 in, represents the first similarity, represents the fused feature vector, represents the knowledge base feature vector, Represents the standard deviation of the preset feature space.

[0036] Step 103: Determine the chat content with a first similarity greater than a preset similarity threshold as the chat content belonging to the knowledge base information, and determine the chat quality evaluation parameter of the first user based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector.

[0037] In some embodiments, when the first similarity is greater than a preset similarity threshold, it can be considered that the chat content corresponding to the first similarity is related to the preset knowledge base information, that is, the chat content is related to learning.

[0038] In some embodiments, the preset similarity threshold may be dynamic, and the value of the preset similarity threshold may be related to the training of the preset recognition model. Specifically, the value of the preset similarity threshold may be determined by the following formula:

[0039] in, represents the preset initial threshold, η represents the maximum adjustment amplitude of the threshold, k represents the adjustment rate of the model, t represents the current training round of the model, and t0 represents the starting round of the model triggering adjustment.

[0040] In some embodiments, the first user's dependency parameters, thinking jump parameters and dynamic penalty parameters can be determined based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector, so as to perform a weighted summation of the dependency parameters, semantic jump distance and dynamic penalty parameters to obtain the first user's chat quality evaluation parameters.

[0041] In other words, in some embodiments, the correlation between the chat information between the first user and the second user and the preset knowledge base information can be determined based on the first number of chat contents, the total number of chat contents, the first similarity, and the fused feature vector, thereby determining whether the chat information between the first user and the second user is related to learning.

[0042] In other words, the chat quality evaluation parameter may indicate whether the chat information between the first user and the second user is related to learning.

[0043] Specifically, in some embodiments, the dependency parameters between the chat information and the preset knowledge base information can be determined based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity, and the dependency parameters are used to indicate the degree of association between the chat information and the preset knowledge base information.

[0044] Specifically, in some embodiments, the semantic jump distance of adjacent chat contents can be determined based on the fused feature vectors in the adjacent chat contents, so as to determine the thinking transition parameters of the chat information based on the semantic jump distance and a preset time decay weight, and the thinking transition parameters are used to indicate the logical relationship between the chat contents.

[0045] Specifically, in some embodiments, a dynamic penalty parameter can be determined based on the first number, the total number and a threshold number of chat contents that do not belong to the preset knowledge base content, and the dynamic penalty parameter is used to indicate the degree of irrelevance of the chat information to the preset knowledge base information.

[0046] Step 104: When the chat quality evaluation parameter is less than the preset chat warning threshold, a warning message is sent to the guardian terminal of the first user, so that the guardian of the first user can determine the learning status of the first user.

[0047] In some embodiments, the larger the chat quality evaluation parameter is, the more relevant the chat information between the first user and the second user is to the preset knowledge base information, that is, the more relevant the chat information between the first user and the second user is to learning.

[0048] In some embodiments, when the chat quality evaluation parameter is less than a preset chat warning threshold, it means that the correlation between the chat information between the first user and the second user and the preset knowledge base information is poor, that is, the chat information between the first user and the second user may not be related to learning. At this time, a warning message can be sent to the guardian terminal of the first user to enable the guardian of the first user to determine the learning status of the first user.

[0049] It should be understood that in some optional embodiments, an early warning message may be sent to the guardian terminal of the second user who is chatting with the first user, so that the guardian of the second user can determine the learning status of the second user. An early warning reminder may also be sent to the first user and / or the second user to prompt the first user and / or the second user to change the learning status and reduce the number of chats not related to learning.

[0050] In some embodiments, the warning information may include: the effective learning time of the first user and / or the second user, wherein the effective learning time may be the learning time corresponding to the chat content belonging to the preset knowledge base information.

[0051] In some optional embodiments, the ratio between the learning time corresponding to the chat content belonging to the preset knowledge base information and the chat time corresponding to the chat information can also be determined as an effective learning parameter; when the effective learning parameter is less than the preset learning efficiency threshold, the learning time is determined as warning information.

[0052] In some embodiments, the warning information further includes at least one of the following: a chat start request of the first user, a chat termination instruction of the first user, a chat start time corresponding to the chat start request, and a chat termination time corresponding to the chat termination instruction.

[0053] It should be understood that the above-mentioned warning information is only an example, and the warning information may also include other information, such as the first user's study subject, the number of times the first user initiates a chat start request within a preset time period, and other information that can be used to indicate the user's learning behavior or user's learning habits.

[0054] In summary, the present disclosure proposes a chat content monitoring method, including: in response to a chat start request and a chat termination instruction initiated by a first user to a second user, determining the chat information between the first user and the second user, and a fused feature vector of one or more chat contents contained in the chat information; based on the fused feature vector, using a preset recognition model, determining a knowledge base feature vector corresponding to the fused feature vector in the preset knowledge base information, so as to determine a first similarity between the chat content and the preset knowledge base information based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space; determining the chat content with a first similarity greater than a preset similarity threshold as the chat content belonging to the knowledge base information, so as to determine the chat quality evaluation parameter of the first user based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector; when the chat quality evaluation parameter is less than the preset chat warning threshold, sending a warning message to the guardian terminal of the first user, so that the guardian of the first user can determine the learning status of the first user. The method disclosed in the present invention extracts the fusion feature vector between the user's chat content, and compares the fusion feature vector with the knowledge base feature vector, so as to judge whether the user's chat content is related to learning, and then evaluate the user's chat quality. When the user's chat quality is less than a preset threshold, it is judged that the user's chat content is poorly related to learning, and it is determined that the user is chatting at this time, and an early warning message is issued, thereby realizing real-time monitoring of the user's learning status, improving the accuracy of monitoring the user's learning status, and being able to timely remind the user and the guardian, thereby improving the user's learning efficiency.

[0055] As a possible implementation, Figure 2 The flowchart of another chat content monitoring method shown in FIG. 1 further explains, based on the above embodiment, the knowledge base feature vector corresponding to the fused feature vector in the preset knowledge base information is determined by using a preset recognition model based on the fused feature vector, and includes the following steps: Step 201: Using a preset feature editor, extract semantic feature vectors and / or formula feature vectors from the chat content.

[0056] In some embodiments, knowledge distillation can be performed on the text in the chat content through a preset feature editor to extract semantic feature vectors in the chat content.

[0057] In some embodiments, a preset feature editor can be used to perform structural analysis on formulas involved in the chat content (for example, analyzing the operation symbols, variable values, etc. used in the formula), thereby extracting formula feature vectors in the chat content.

[0058] In some embodiments, the preset feature editor is, for example, a bidirectional feature encoder (Bidirectional Encoder Representation from Transformers, BERT), or a variant feature editor of BERT, etc., which is not limited in the present disclosure.

[0059] It should be understood that when extracting feature vectors from chat content, only semantic feature vectors may be extracted, or only formula feature vectors may be extracted, or both semantic feature vectors and formula feature vectors may be extracted.

[0060] For example, when the chat content contains only text information, only semantic feature vectors can be extracted; when the chat content contains only formula information, only formula feature vectors can be extracted; when the chat content contains both text information and formula information, both semantic feature vectors and formula feature vectors can be extracted.

[0061] For example, when the chat content is: The formula of the Pythagorean theorem is , then the semantic feature vector can be extracted according to the "Pythagorean Theorem", and the semantic feature vector can be extracted according to the " ” Extract the formula feature vector.

[0062] Step 202: Perform weighted summation on the semantic feature vectors and / or formula feature vectors to determine a fused feature vector of the chat content.

[0063] In some embodiments, a weighted summation may be performed on the semantic feature vector and the formula feature vector to achieve feature fusion, thereby determining a fused feature vector of the chat content, wherein the weights corresponding to the semantic feature vector and the formula feature vector are not limited.

[0064] For example, with the formula , determine the fusion feature vector, where represents the fused feature vector, represents the semantic feature vector, represents the formula eigenvector.

[0065] In some embodiments, when only a semantic feature vector is extracted from the chat content, the semantic feature vector can be directly determined as a fused feature vector; when only a formula feature vector is extracted from the chat content, the formula feature vector can be directly determined as a fused feature vector.

[0066] In summary, the present disclosure proposes a chat content monitoring method, including: using a preset feature editor to extract semantic feature vectors and / or formula feature vectors in the chat content; performing weighted summation on the semantic feature vectors and / or formula feature vectors to determine a fused feature vector of the chat content. The present disclosure generates a fused feature vector by extracting semantic feature vectors and / or formula feature vectors in the chat content, thereby improving the accuracy of determining feature extraction, and further improving the accuracy of determining whether the chat content belongs to knowledge base information.

[0067] As a possible implementation, Figure 3 The flowchart of another chat content monitoring method shown in the figure shows a method for obtaining a preset recognition model based on the above embodiment, including the following steps: Step 301: parse the preset knowledge base information to obtain a training data set.

[0068] In some embodiments, the text information in the preset knowledge base can be parsed into knowledge base text segments based on information such as chapters in the textbook. Furthermore, the text segment can be a tag sequence marked with a specific level.

[0069] In some embodiments, the formulas in the preset knowledge base can be parsed into a coding format (such as LaTeX format), and then specific formula information with a graph structure can be generated according to the syntax corresponding to the coding, where the node set in the graph structure can be the operators, operands, special symbols, etc. in the formula, and the edge set in the graph structure can be the dependency relationship between the corresponding coding syntax in the formula.

[0070] In some embodiments, coordinate annotation may be performed on a chart in a preset knowledge base, thereby parsing the image into a chart image with coordinate annotations.

[0071] In other words, the preset knowledge base information can be parsed and the parsing result can be determined as a training data set. In other words, the training data set can include at least one of a knowledge base text segment, formula information with a syntax diagram structure, and a chart image with coordinate annotations.

[0072] Step 302: randomly replace nodes in the syntax graph structure to obtain replacement formula information, and randomly change the positions of coordinate annotations to obtain a disturbed graph image.

[0073] In some embodiments, in order to improve the recognition accuracy of a preset recognition model, the data in the training data set can be divided into positive samples and negative samples, so that the recognition model to be trained can better learn the correct sample information during training, thereby improving the model learning efficiency.

[0074] In some embodiments, nodes in the syntax graph structure may be randomly replaced to replace the parsed formula information and generate erroneous formula information, namely, replacement formula information.

[0075] In some embodiments, the position of the coordinate annotations may be changed to generate an erroneous chart image, ie, to disrupt the chart image.

[0076] Among them, the coordinate annotation of the chart can be changed by the following formula:

[0077] in, represents the perturbed chart image, I represents the chart image, and Δ represents the displacement of the coordinate annotation.

[0078] Step 303: taking the knowledge base text segment, formula information and chart image as positive training samples, and taking the replaced formula information and disturbed chart image as negative training samples, and performing first training on the recognition model to be trained to obtain a first training recognition model.

[0079] In some embodiments, the determined knowledge base text segments, formula information and chart images (i.e., the training data set obtained by parsing the preset knowledge base information) are used as positive training samples, and the replaced formula information and disturbed chart images are used as negative training samples, and the first training is performed on the to-be-trained recognition model, so that the to-be-trained model can accelerate the learning of the correct preset knowledge base information.

[0080] In some embodiments, the loss function used in the first training may be as follows, but is not limited thereto:

[0081] in, represents the loss function, a represents the preset anchor sample as a reference, p represents the positive training sample, and n represents the negative training sample.

[0082] Step 304: Perform synonym replacement and / or voice conversion on the knowledge base text segment to obtain a replacement text segment; and randomly discard nodes in the grammar graph structure to obtain missing formula information.

[0083] In some embodiments, in order to enhance the recognition capability of the preset recognition model between preset knowledge base information and non-preset knowledge base information, the training data set may be fine-tuned.

[0084] In some embodiments, synonym replacement and / or voice conversion are performed on the knowledge base text segment to provide richer semantic training samples.

[0085] Among them, the formula used for synonym replacement can be as follows:

[0086] in, represents synonym replacement, x is the word to be replaced, w is the synonym of x, and cos represents the similarity between x and w.

[0087] In some examples, nodes in the syntax graph structure may be randomly discarded to obtain missing formula information, thereby providing richer formula information.

[0088] Among them, the random discarding of nodes in the syntax graph structure can be achieved by the following formula:

[0089] in, Indicates missing formula information, G indicates formula information, DropNode indicates a random drop function, and p indicates a preset parameter in the function.

[0090] Step 305: Based on the replacement text segment, the missing formula information and the disturbed chart image, the first training recognition model is subjected to second training to obtain a preset recognition model.

[0091] In some embodiments, the first training recognition model may be subjected to second training based on the replacement text segment, the missing formula information, and the disturbed chart image to improve the recognition capability of the preset recognition model obtained after the training.

[0092] In some embodiments, the second training may include multiple stages of training, such as training at elementary, intermediate, and advanced stages, wherein the sample complexity may be gradually increased at each stage, thereby improving the accuracy of recognition by the preset recognition model.

[0093] In some embodiments, when performing the first training and / or the second training, the missing formula information, the replacement formula information, the formula information, etc. may be convolutionally iterated and updated by the following formula to improve the recognition accuracy of the preset recognition model:

[0094] in, Represents the feature vector of node v in the graph structure at the kth convolutional layer of the model, represents the neighbor set of node v, Represents the normalized coefficient of the edge (v,u) between node v and node u in the graph structure. represents the learnable weight matrix of the kth layer, Represents the activation function.

[0095] In some embodiments, when performing the first training and / or the second training, the replacement text segment, the knowledge base text segment, etc. may be encoded by the following formula to facilitate the learning of the model:

[0096] Among them, PE represents the encoding result, c represents the chapter number, indicating the chapter position of the text segment (i.e., the replacement text segment and / or the knowledge base text segment) in the textbook, l represents the hierarchical depth of the text segment (for example, if the text segment belongs to a chapter, l is 1, if the text segment belongs to a section, l is 2, and if the text segment belongs to a subsection, l is 3), pos represents the position of the character in the paragraph, A represents the hidden dimension of the model, i represents the dimension index, The term is used to realize the model's attenuated attention to deep chapter structures. The deeper the level (such as subsections), the smaller the fluctuation of position encoding.

[0097] In summary, the chat content monitoring method proposed in the present invention improves the recognition ability of the preset recognition model obtained using the training samples by dividing the training samples into positive and negative samples and partially modifying or replacing the samples, thereby further improving the accuracy of judging whether the student chat content is related to the knowledge base information.

[0098] As a possible implementation, Figure 4 The flowchart of another chat content monitoring method shown in FIG. 1 further explains step 103 based on the above embodiment, and includes the following steps: Step 401, based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity, determine the dependency parameters between the chat information and the preset knowledge base information.

[0099] In some embodiments, the dependency parameter may be determined based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity, in combination with the following formula 2: Formula 2, Where DI represents the dependency parameter, K represents the first number of chat contents belonging to the preset knowledge base information, and N represents the total number of chat contents. Indicates the first similarity of the i-th chat content.

[0100] In some embodiments, the dependency parameter is used to indicate the degree of association between the chat information and the preset knowledge base information.

[0101] Step 402, based on the fused feature vectors in the adjacent chat contents, the semantic jump distance of the adjacent chat contents is determined, so as to determine the thought transition parameter of the chat information based on the semantic jump distance and the preset time decay weight.

[0102] In some embodiments, based on the fused feature vectors in the adjacent chat contents, the semantic jump distance of the adjacent chat contents is determined in combination with the following formula 3: Formula 3, in, represents the fused feature vector of the jth chat content, Indicates the jth chat content Chat content with Article j-1 The semantic jump distance between them; Then, based on the semantic jump distance and the preset time decay weight, combined with the following formula 4, the thought transition parameter of the chat information is determined: Formula 4, Among them, CLS represents the thinking transition parameter, Indicates the semantic jump distance in the chat message The mean of Indicates the semantic jump distance in the chat message , Indicates the preset time decay weight.

[0103] In some embodiments, the thought transition parameter is used to indicate the logical relationship between the chat contents. Step 403, determining a dynamic penalty parameter based on the first number, the total number of items, and a threshold number of items of chat content that do not belong to the preset knowledge base content.

[0104] In some embodiments, the second number of chat contents in the chat information that do not belong to the preset knowledge base can be determined by subtracting the first number and the total number; and then based on the second number, the total number and the number threshold of chat contents that do not belong to the preset knowledge base content, combined with the following formula 5, the dynamic penalty parameter is determined: Formula 5, in, represents the dynamic penalty parameter, Represents the second number, Indicates the number threshold.

[0105] In some embodiments, the dynamic penalty parameter is used to indicate the degree of irrelevance of the chat information to the preset knowledge base information.

[0106] Step 404: Perform weighted summation on the dependency parameter, the semantic jump distance, and the dynamic penalty parameter to obtain a chat quality evaluation parameter of the first user.

[0107] In some examples, a weighted sum is performed on the dependency parameter, the semantic jump distance, and the dynamic penalty parameter, so that the weighted sum result is determined as the chat quality evaluation parameter of the first user.

[0108] In summary, the chat content monitoring method proposed in the present invention determines the dependency parameters, semantic jump distance and dynamic penalty parameters to determine the chat quality evaluation parameters according to the dependency parameters, semantic jump distance and dynamic penalty parameters, so as to evaluate whether the user's chat content is related to the preset learning knowledge base from multiple angles, thereby improving the objectivity and accuracy of the chat quality evaluation parameters and reducing the probability of false triggering of warning information.

[0109] The following is an exemplary description of a chat content monitoring method proposed in the present disclosure. Figure 5 The steps shown may include: 1. Perform multimodal extraction on the chat information input by the user to obtain semantic feature vectors and / or formula feature vectors.

[0110] 2. Perform knowledge distillation on the knowledge base features (i.e., the preset knowledge base information mentioned above) and cache them to increase the rate of comparing chat information and preset knowledge base information.

[0111] 3. Perform multimodal extraction and cross-modal fusion on the cached information to obtain the knowledge base feature vector.

[0112] 4. Perform cross-modal fusion on the semantic feature vector and / or the formula feature vector to obtain a fused feature vector.

[0113] 5. Compare the similarity between the knowledge base feature vector and the fusion feature vector to dynamically determine whether the chat content is time-domain knowledge base information.

[0114] Corresponding to the above-mentioned chat content monitoring method, the present invention also provides a chat content monitoring device. Since the device embodiment of the present invention corresponds to the above-mentioned method embodiment, the details not disclosed in the device embodiment can be referred to the above-mentioned method embodiment, and will not be repeated in the present invention.

[0115] Figure 6 A schematic diagram of the structure of a chat content monitoring device provided by an embodiment of the present disclosure, such as Figure 6 As shown, the device comprises: A first determining unit 610 is configured to determine, in response to a chat start request and a chat termination instruction initiated by the first user to the second user, chat information between the first user and the second user, and a fusion feature vector of one or more chat contents contained in the chat information; A second determining unit 620 is used to determine, based on the fused feature vector and using a preset recognition model, a knowledge base feature vector corresponding to the fused feature vector in the preset knowledge base information, so as to determine a first similarity between the chat content and the preset knowledge base information based on the fused feature vector, the knowledge base feature vector, and a standard deviation of the preset feature space; A third determining unit 630 is used to determine the chat content with a first similarity greater than a preset similarity threshold as the chat content belonging to the knowledge base information, so as to determine the chat quality evaluation parameter of the first user based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector; The warning unit 640 is used to send a warning message to the guardian terminal of the first user when the chat quality evaluation parameter is less than a preset chat warning threshold, so that the guardian of the first user can determine the learning status of the first user.

[0116] In some embodiments of the present disclosure, the first determination unit 610 is further used to: use a preset feature editor to extract semantic feature vectors and / or formula feature vectors in the chat content; and perform weighted summation of the semantic feature vectors and / or formula feature vectors to determine a fused feature vector of the chat content.

[0117] In some embodiments of the present disclosure, the first determination unit 610 is further used to: parse the preset knowledge base information to obtain a training data set, the training data set including at least one of the following: a knowledge base text segment, formula information with a grammatical graph structure, and a chart image with coordinate annotations; randomly replace nodes in the grammatical graph structure to obtain replacement formula information, and randomly change the position of the coordinate annotations to obtain a disturbed chart image; use the knowledge base text segment, formula information, and chart image as positive training samples, and use the replacement formula information and the disturbed chart image as negative training samples to perform a first training on the recognition model to be trained to obtain a first training recognition model; perform synonym replacement and / or voice conversion on the knowledge base text segment to obtain a replacement text segment; and randomly discard nodes in the grammatical graph structure to obtain missing formula information; based on the replacement text segment, the missing formula information, and the disturbed chart image, perform a second training on the first training recognition model to obtain a preset recognition model.

[0118] In some embodiments of the present disclosure, the second determining unit 620 is further configured to determine the first similarity based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space in combination with the following formula 1: Formula 1, in, represents the first similarity, represents the fused feature vector, represents the knowledge base feature vector, Represents the standard deviation of the preset feature space.

[0119] In some embodiments of the present disclosure, the third determination unit 630 is further used to: determine a dependency parameter between the chat information and the preset knowledge base information based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity, the dependency parameter being used to indicate the degree of association between the chat information and the preset knowledge base information; determine the semantic jump distance of adjacent chat contents based on the fused feature vectors in the adjacent chat contents, and determine the thinking jump parameter of the chat information based on the semantic jump distance and the preset time attenuation weight, the thinking jump parameter being used to indicate the logical relationship between the chat contents; determine a dynamic penalty parameter based on the first number, the total number, and the threshold number of chat contents that do not belong to the preset knowledge base content, the dynamic penalty parameter being used to indicate the degree of irrelevance of the chat information to the preset knowledge base information; and perform weighted summation of the dependency parameter, the semantic jump distance, and the dynamic penalty parameter to obtain a chat quality evaluation parameter of the first user.

[0120] In some embodiments of the present disclosure, the third determining unit 630 is further configured to determine the dependency parameter based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity, in combination with the following formula 2: Formula 2, Where DI represents the dependency parameter, K represents the first number of chat contents belonging to the preset knowledge base information, and N represents the total number of chat contents. Indicates the first similarity of the i-th chat content.

[0121] In some embodiments of the present disclosure, the third determination unit 630 is further configured to: determine the dynamic penalty parameter based on the first number, the total number, and the threshold number of chat contents that do not belong to the preset knowledge base content, including: determining the second number of chat contents that do not belong to the preset knowledge base in the chat information based on the first number and the total number; and determining the dynamic penalty parameter based on the second number, the total number, and the threshold number of chat contents that do not belong to the preset knowledge base content, in combination with the following formula 5: Formula 5, in, represents the dynamic penalty parameter, Represents the second number, Indicates the number threshold.

[0122] In some embodiments of the present disclosure, the warning determination unit 640 is also used to: determine the ratio between the learning time corresponding to the chat content belonging to the preset knowledge base information and the chat time corresponding to the chat information as an effective learning parameter; when the effective learning parameter is less than the preset learning efficiency threshold, determine the learning time as warning information.

[0123] In some embodiments of the present disclosure, the warning information also includes at least one of the following: a chat start request of the first user, a chat termination instruction of the first user, a chat start time corresponding to the chat start request, and a chat termination time corresponding to the chat termination instruction.

[0124] It should be noted that the above explanation of the method embodiment is also applicable to the device of this embodiment, and the principle is the same, which is not limited in this embodiment.

[0125] Based on the above Figures 1 to 5 The method shown in the figure, accordingly, this embodiment also provides a computer program product, including a computer program, which implements the above-mentioned Figures 1 to 5 The method shown.

[0126] Based on the above Figures 1 to 5 The method shown in the embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned Figures 1 to 5 The method shown.

[0127] Based on this understanding, the technical solution 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, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present application.

[0128] like Figure 7 The figure shows a hardware structure diagram of an electronic device of the present invention, including: at least one processor 701; and, A memory 702 that is communicatively connected to at least one of the processors 701; wherein, The memory 702 stores instructions that can be executed by at least one of the processors, and the instructions are executed by at least one of the processors so that at least one of the processors can execute a chat content monitoring method as described above.

[0129] Figure 7 A processor 701 is taken as an example.

[0130] The electronic device may further include: an input device 703 and a display device 704 .

[0131] The processor 701, the memory 702, the input device 703 and the display device 704 may be connected via a bus or other means, and the figure takes the connection via a bus as an example.

[0132] The memory 702 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the review content generation method in the embodiment of the present application, for example, Figures 1 to 5 The processor 701 executes various functional applications and data processing by running the non-volatile software programs, instructions and modules stored in the memory 702, that is, a chat content monitoring method in the above embodiment is implemented.

[0133] The memory 702 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required by at least one function; the data storage area may store data created according to the use of the review content generation method, etc. In addition, the memory 702 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 702 may optionally include a memory remotely arranged relative to the processor 701, and these remote memories may be connected to the device for executing the review content generation method via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0134] The input device 703 can receive user clicks and generate signal inputs related to user settings and function controls of the review content generation method. The display device 704 can include display devices such as a display screen.

[0135] The one or more modules are stored in the memory 702, and when executed by the one or more processors 701, a chat content monitoring method in any of the above method embodiments is executed.

[0136] Optionally, the above-mentioned physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a WI-FI interface), etc.

[0137] Those skilled in the art will appreciate that the above-mentioned physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different arrangements of components.

[0138] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the above-mentioned physical device, and supports the operation of the information processing program and other software and / or programs. The network communication module is used to realize the communication between the components inside the storage medium, and the communication with other hardware and software in the information processing physical device.

[0139] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform, or by hardware.

[0140] By applying the solution of this embodiment, compared with the current existing technology, this solution extracts the fusion feature vector between the user's chat content, and uses the fusion feature vector to compare with the knowledge base feature vector, so as to determine whether the user's chat content is related to learning, and then evaluate the user's chat quality. When the user's chat quality is less than a preset threshold, it is determined that the user's chat content is poorly relevant to learning, and it is determined that the user is chatting at this time, and an early warning message is issued, thereby realizing real-time monitoring of the user's learning status, improving the accuracy of monitoring the user's learning status, and being able to timely remind users and guardians, thereby improving the user's learning efficiency.

[0141] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0142] The above is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments described herein, but will conform to the widest scope consistent with the principles and novel features applied for herein.

Claims

1. A chat content monitoring method, characterized in that: The method comprises: In response to a chat start request and a chat termination instruction initiated by a first user to a second user, determining chat information between the first user and the second user, and a fusion feature vector of one or more chat contents contained in the chat information; Based on the fused feature vector, using a preset recognition model, determining a knowledge base feature vector in the preset knowledge base information corresponding to the fused feature vector, so as to determine a first similarity between the chat content and the preset knowledge base information based on the fused feature vector, the knowledge base feature vector, and a standard deviation of a preset feature space; Determine the chat content whose first similarity is greater than a preset similarity threshold as the chat content belonging to the knowledge base information, and determine the chat quality evaluation parameter of the first user based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector; When the chat quality evaluation parameter is less than a preset chat warning threshold, a warning message is sent to a guardian terminal of the first user, so that the guardian of the first user can determine the learning status of the first user.

2. The method according to claim 1, characterized in that: Determining a fusion feature vector of one or more chat contents contained in the chat information includes: Using a preset feature editor, extracting semantic feature vectors and / or formula feature vectors from the chat content; A weighted sum is performed on the semantic feature vectors and / or formula feature vectors to determine a fusion feature vector of the chat content.

3. The method according to claim 1, characterized in that: The method further comprises: Parsing the preset knowledge base information to obtain a training data set, wherein the training data set includes at least one of the following: a knowledge base text segment, formula information with a syntax diagram structure, and a chart image with coordinate annotations; Randomly replacing nodes in the syntax graph structure to obtain replacement formula information, and randomly changing the positions of the coordinate annotations to obtain a disturbed graph image; Using the knowledge base text segment, the formula information and the chart image as positive training samples, and using the replacement formula information and the disturbed chart image as negative training samples, a first training is performed on the training recognition model to obtain a first training recognition model; Perform synonym replacement and / or voice conversion on the knowledge base text segment to obtain a replacement text segment; and randomly discard nodes in the grammar graph structure to obtain missing formula information; Based on the replacement text segment, the missing formula information and the disturbed chart image, the first training recognition model is subjected to second training to obtain the preset recognition model.

4. The method according to claim 1, characterized in that: The determining, based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space, the first similarity between the chat content and the preset knowledge base information comprises: Based on the fused feature vector, the knowledge base feature vector, and the standard deviation of the preset feature space, the first similarity is determined in combination with the following formula 1: Formula 1, in, represents the first similarity, represents the fused feature vector, represents the knowledge base feature vector, Represents the standard deviation of the preset feature space.

5. The method according to claim 1, characterized in that: The determining of the chat quality evaluation parameter of the first user based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, the first similarity, and the fused feature vector comprises: Determine, based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity, a dependency parameter between the chat information and the preset knowledge base information, wherein the dependency parameter is used to indicate a degree of association between the chat information and the preset knowledge base information; Determine the semantic jump distance of the adjacent chat contents based on the fused feature vectors in the adjacent chat contents, and determine the thought transition parameter of the chat information based on the semantic jump distance and a preset time decay weight, wherein the thought transition parameter is used to indicate the logical relationship between the chat contents; Determine a dynamic penalty parameter based on the first number of messages, the total number of messages, and a threshold number of messages of the chat content that do not belong to the preset knowledge base content, wherein the dynamic penalty parameter is used to indicate the degree of irrelevance of the chat information to the preset knowledge base information; A weighted sum is performed on the dependency parameter, the semantic jump distance, and the dynamic penalty parameter to obtain a chat quality evaluation parameter of the first user.

6. The method according to claim 5, characterized in that: The determining of the dependency parameter between the chat information and the preset knowledge base information based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity includes: The dependency parameter is determined based on the first number of chat contents belonging to the preset knowledge base information, the total number of chat contents, and the first similarity in combination with the following formula 2: Formula 2, Wherein, DI represents the dependency parameter, K represents the first number of chat contents belonging to the preset knowledge base information, and N represents the total number of chat contents. Indicates the first similarity of the i-th chat content.

7. The method according to claim 5, characterized in that: The determining of the semantic jump distance of the adjacent chat contents based on the fused feature vectors in the adjacent chat contents, and determining the thought transition parameter of the chat information based on the semantic jump distance and a preset time decay weight includes: Based on the fused feature vectors in the adjacent chat contents, combined with the following formula 3, the semantic jump distance of the adjacent chat contents is determined: Formula 3, in, represents the fused feature vector of the jth chat content, Indicates the jth chat content Chat content with Article j-1 The semantic jump distance between them; Based on the semantic jump distance and the preset time decay weight, combined with the following formula 4, the thought transition parameter of the chat information is determined: Formula 4, Wherein, CLS represents the thought transition parameter, Indicates the semantic jump distance in the chat message The mean of Indicates the semantic jump distance in the chat message The standard deviation of Indicates the preset time decay weight.

8. The method according to claim 5, characterized in that: The determining of the dynamic penalty parameter based on the first number of chats, the total number of chats, and the number threshold of chats that do not belong to the preset knowledge base content includes: Based on the first number and the total number, determining a second number of chat contents in the chat information that do not belong to the preset knowledge base; Based on the second number of chats, the total number of chats, and the number threshold of chats that do not belong to the preset knowledge base content, the dynamic penalty parameter is determined in combination with the following formula 5: Formula 5, in, represents the dynamic penalty parameter, the represents the second number, the Indicates the number threshold.

9. The method according to claim 1 or 5, characterized in that: The method further comprises: Determine the ratio between the learning time corresponding to the chat content belonging to the preset knowledge base information and the chat time corresponding to the chat information as an effective learning parameter; When the effective learning parameter is less than a preset learning efficiency threshold, the learning duration is determined as warning information.

10. The method according to claim 1, characterized in that The warning information further includes at least one of the following: a chat start request of the first user, a chat termination instruction of the first user, a chat start time corresponding to the chat start request, and a chat termination time corresponding to the chat termination instruction.

Images