Thematic map knowledge intelligent recommendation method and device, and electronic equipment

Abstract

The application discloses a thematic map knowledge intelligent recommendation method and device and electronic equipment, and relates to the technical field of thematic maps. The method comprises the following steps: constructing a multi-dimensional feature matrix according to multi-dimensional information of a thematic map; analyzing the dynamic demand of a target user according to real-time interaction behavior data of the target user and the thematic map and the multi-dimensional feature matrix, obtaining a real-time demand vector of the target user, matching each piece of knowledge in a preset knowledge base of the thematic map in combination with the multi-dimensional feature matrix, obtaining matching parameters of each piece of knowledge for the target user, and determining and pushing target knowledge accordingly. The method replaces traditional single-dimensional information with a multi-dimensional matrix, realizes dynamic judgment of the matching degree between each piece of information in the knowledge base and the real-time demand of the user through the multi-dimensional feature matrix, screens out knowledge matched with the real-time demand of the user for pushing, thereby improving the accuracy and diversity of the thematic map knowledge recommendation, and enhancing the dynamic adaptability.

Description

Technical Field

[0001] This application relates to the field of thematic map technology, and in particular to a method, apparatus and electronic device for intelligent recommendation of thematic map knowledge. Background Technology

[0002] Thematic maps, also known as special maps, are maps that, on a geographic base map, highlight and comprehensively represent one or more knowledge elements related to the theme according to the requirements of the map's subject matter, thus making the map content thematic, the expression form diverse, and the purpose specialized.

[0003] Current thematic map knowledge recommendation schemes typically push knowledge content corresponding to the tags with the highest matching degree based on the search keywords entered by the user in real time, or push similar knowledge content based on the user's historical search preferences, historical browsing preferences, and other data.

[0004] However, limited by a single matching dimension such as search keywords or user historical data, the knowledge push results of thematic maps are often too concentrated on one type of knowledge, which cannot meet users' needs for understanding multiple aspects of knowledge related to thematic maps. Moreover, when users' needs change during the use of thematic maps, the recommended content often lags behind the user's real-time intentions. It can be seen that the current thematic map knowledge recommendation scheme has insufficient accuracy and diversity, and poor dynamic adaptability. Summary of the Invention

[0005] The main objective of this application is to propose a method, apparatus, and electronic device for intelligent recommendation of thematic map knowledge, aiming to improve the accuracy and diversity of thematic map knowledge recommendation and enhance its dynamic adaptability.

[0006] In a first aspect, the present invention provides a method for intelligent recommendation of thematic map knowledge, comprising: Based on the multi-dimensional information of the thematic map, a multi-dimensional feature matrix of the thematic map is constructed; Based on the real-time interaction data between the target user and the thematic map, and the multi-dimensional feature matrix, a preset demand analysis model is used to dynamically analyze the target user's demand for the thematic map, thereby obtaining the target user's real-time demand vector for the thematic map. Based on the real-time demand vector and the multi-dimensional feature matrix, a preset knowledge matching model is used to match each piece of knowledge in the preset knowledge base of the thematic map to obtain the matching parameters of the target user for each piece of knowledge. Based on the matching parameters of each piece of knowledge, the target knowledge is determined from each piece of knowledge in the preset knowledge base, and the target knowledge is pushed to the target user.

[0007] In an optional implementation, before constructing the multidimensional feature matrix of the thematic map based on its multidimensional information, the method further includes: Extract the thematic dimension information of the thematic map from its metadata; Extract the knowledge attribute dimension information of the thematic map from the knowledge attribute information corresponding to the thematic map; The user feature dimension information of the thematic map is extracted from the user feature information corresponding to the thematic map. The multi-dimensional information includes: the theme dimension information, the knowledge attribute dimension information, and the user feature dimension information.

[0008] In an optional implementation, the user characteristic dimension information includes: user knowledge level information, usage scenario information, historical interaction behavior information, and historical feedback information; The step involves using a preset demand analysis model to dynamically analyze the target user's demands based on real-time interaction data between the target user and the thematic map, and the multi-dimensional feature matrix, to obtain the target user's real-time demand vector for the thematic map, including: Based on the real-time interaction behavior data, and the historical interaction behavior information and historical feedback information in the user feature dimension information of the multi-dimensional feature matrix, the demand parsing model is used to dynamically parse the target user's demand to obtain the real-time demand vector.

[0009] In an optional implementation, before obtaining the real-time demand vector by dynamically analyzing the target user's demands using the demand analysis model based on the real-time interaction behavior data and the historical interaction behavior information and historical feedback information in the user feature dimension information of the multi-dimensional feature matrix, the method further includes: Obtain user thematic map interaction samples, wherein the user thematic map interaction samples include: real-time interaction behavior data of the samples, corresponding historical interaction behavior information and historical feedback information of the samples, and pre-labeled sample real needs; The demand analysis model is obtained by training the model based on the user thematic map interaction samples.

[0010] In an optional implementation, the knowledge matching model includes: an attention weighting module and a knowledge matching module; the step of matching each piece of knowledge in the preset knowledge base of the thematic map using a preset knowledge matching model based on the real-time demand vector and the multi-dimensional feature matrix to obtain the matching parameters of the target user for each piece of knowledge includes: Using the attention weight module, dynamic weights are calculated for the topic dimension information and the knowledge attribute dimension information in the multidimensional feature matrix according to the real-time demand vector, to obtain the first attention weight vector of the topic dimension and the second attention weight vector of the knowledge attribute dimension. The knowledge matching module extracts the corresponding topic feature vector and knowledge attribute feature vector from each piece of knowledge in the preset knowledge base, and determines the matching parameters of each piece of knowledge based on the first attention weight vector, the second attention weight vector, the topic feature vector, and the knowledge attribute feature vector.

[0011] In an optional implementation, determining the target knowledge from each piece of knowledge in the preset knowledge base based on the matching parameters of each piece of knowledge includes: Based on the matching parameters of each piece of knowledge, the pieces of knowledge are sorted in descending order, and the first preset number of pieces of knowledge are selected as multiple candidate pieces of knowledge. Based on preset knowledge constraints, the multiple candidate knowledge items are optimized to obtain optimized knowledge. The target knowledge is obtained from the optimized knowledge.

[0012] In an optional implementation, the preset knowledge constraints include: type constraints; The optimization of the multiple candidate knowledge items based on preset knowledge constraints to obtain optimized knowledge includes: Based on the type constraints, statistically analyze the knowledge type distribution of the multiple candidate knowledge items; Based on the knowledge type distribution, determine whether there is a target knowledge type among the multiple candidate knowledge items whose proportion exceeds a preset proportion threshold; If the target knowledge type exists among the multiple candidate knowledge items, then the knowledge item with the lowest matching parameter among the target knowledge types is selected, and the unselected knowledge items with the highest matching parameters among other knowledge types are added to obtain the optimized knowledge.

[0013] In an optional implementation, the preset knowledge constraint further includes: a correlation constraint. The step of optimizing the multiple candidate knowledge items according to preset knowledge constraints to obtain optimized knowledge further includes: Based on the aforementioned correlation constraint, calculate the correlation between any two pieces of knowledge among the multiple candidate knowledge items; If at least two of the candidate knowledge items have a correlation degree exceeding a preset correlation degree threshold, the knowledge item with the lowest matching parameter among the at least two candidate knowledge items is removed to obtain the optimized knowledge.

[0014] Secondly, the present invention provides a thematic map knowledge intelligent recommendation device, comprising: The construction module is used to construct a multi-dimensional feature matrix of the thematic map based on the multi-dimensional information of the thematic map; The parsing module is used to perform dynamic demand parsing on the target user based on the real-time interaction behavior data between the target user and the thematic map, as well as the multi-dimensional feature matrix, using a preset demand parsing model to obtain the real-time demand vector of the target user for the thematic map. The matching module is used to match each piece of knowledge in the preset knowledge base of the thematic map according to the real-time demand vector and the multi-dimensional feature matrix, using a preset knowledge matching model, to obtain the matching parameters of the target user for each piece of knowledge; The filtering module is used to determine target knowledge from each piece of knowledge in the preset knowledge base based on the matching parameters of each piece of knowledge, and push the target knowledge to the target user.

[0015] Thirdly, the present invention provides an electronic device, comprising: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of any of the methods described in the foregoing embodiments.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method as described in any of the foregoing embodiments.

[0017] The beneficial effects of this application are: The thematic map knowledge intelligent recommendation method provided in this application includes: constructing a multi-dimensional feature matrix of the thematic map based on multi-dimensional information of the thematic map; performing dynamic demand analysis on the target user using a preset demand analysis model based on real-time interaction behavior data between the target user and the thematic map, and the multi-dimensional feature matrix, to obtain a real-time demand vector of the target user for the thematic map; matching each piece of knowledge in a preset knowledge base of the thematic map using a preset knowledge matching model based on the real-time demand vector and the multi-dimensional feature matrix, to obtain matching parameters of the target user for each piece of knowledge; determining target knowledge from each piece of knowledge in the preset knowledge base based on the matching parameters of each piece of knowledge, and pushing the target knowledge to the target user. In this embodiment, by collecting multi-dimensional information from the thematic map and establishing a multi-dimensional matrix based on the collected multi-dimensional information to replace the single-dimensional information in the traditional solution, it is possible to dynamically obtain an implementation demand vector representing the user's real-time needs after each real-time user interaction by combining a preset demand analysis model with the multi-dimensional feature matrix. Then, by combining a preset knowledge matching model with the implementation demand vector and the multi-dimensional feature matrix, the matching degree between each piece of information in the preset knowledge base of the thematic map and the user's real-time needs is judged, so as to filter out the knowledge that matches the user's real-time needs and push it, thereby improving the accuracy and diversity of thematic map knowledge recommendation and enhancing the dynamic adaptability of thematic map knowledge recommendation. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0019] Figure 1 A schematic flowchart of a thematic map knowledge intelligent recommendation method provided in an embodiment of this application; Figure 2 A schematic flowchart of a thematic map knowledge intelligent recommendation method provided in another embodiment of this application; Figure 3 A schematic flowchart of a thematic map knowledge intelligent recommendation method provided in another embodiment of this application; Figure 4 A schematic diagram of the structure of a thematic map knowledge intelligent recommendation device provided in this application embodiment; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0023] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0024] Current thematic map knowledge recommendation schemes typically push knowledge content corresponding to the most relevant tags based on the user's real-time input search keywords, or push similar knowledge content based on the user's historical search preferences and browsing history. However, limited by a single matching dimension such as search keywords or user historical data, the thematic map knowledge push results are often too concentrated on one type of knowledge, failing to meet users' needs for understanding diverse aspects related to the thematic map. Furthermore, when users' needs change during the use of the thematic map, because the proportion of data related to historical search preferences and browsing history is far greater than the proportion of data related to the user's new needs, the recommended content often still corresponds to the knowledge data related to the user's historical search preferences, lagging behind the user's real-time intent. It is evident that current thematic map knowledge recommendation schemes have insufficient accuracy and diversity in knowledge recommendation, and poor dynamic adaptability.

[0025] To address the aforementioned issues, the main objective of this application is to propose an intelligent recommendation method for thematic map knowledge, aiming to improve the accuracy and diversity of thematic map knowledge recommendations and enhance their dynamic adaptability.

[0026] Figure 1 This is a schematic flowchart of a thematic map knowledge intelligent recommendation method provided in an embodiment of this application. The executing entity of this method can be, for example, a computer, a server, or other device with computing power, but is not limited thereto. Figure 1 As shown, the method includes: S101. Based on the multi-dimensional information of the thematic map, construct the multi-dimensional feature matrix of the thematic map.

[0027] For example, the multi-dimensional information of the thematic map mentioned above may be collected and extracted from the thematic map. The dimensions of the multi-dimensional information of the thematic map may include, but are not limited to, dimensions such as theme, knowledge attributes, and user characteristics.

[0028] Taking the example that the multi-dimensional information of the aforementioned thematic map includes three dimensions: theme, knowledge attributes, and user characteristics, the multi-dimensional feature matrix for constructing the thematic map can, for example, refer to constructing a three-dimensional feature matrix corresponding to the thematic map. This three-dimensional feature matrix can be represented, for example, in the following form: M∈R^(T_dim×A_dim×U_dim), In this context, M can represent the three-dimensional feature matrix, R can represent a vector sequence, T_dim can represent the total number of features in the topic dimension, A_dim can represent the total number of features in the knowledge attribute dimension, U_dim can represent the total number of features in the user feature dimension, and R^(T_dim×A_dim×U_dim) means that R is a set of T_dim×A_dim×U_dim dimensional vectors.

[0029] For example, assuming the total number of features in the topic dimension is 25, the total number of features in the knowledge attribute dimension is 7, and the total number of features in the user feature dimension is 7, then M∈R^(T_dim×A_dim×U_dim)=M∈R^(25×7×7)=M∈R^1225. It should be understood that the above is only a possible example, and the actual representation of the multidimensional feature matrix and the dimensions of the vector set to which the multidimensional feature matrix belongs are not limited to the three-dimensional feature matrix and M∈R^1225 mentioned above.

[0030] S102. Based on the real-time interaction data between the target user and the aforementioned thematic map, and the aforementioned multi-dimensional feature matrix, a preset demand analysis model is used to perform dynamic demand analysis on the target user to obtain the real-time demand vector of the target user for the aforementioned thematic map.

[0031] For example, the aforementioned pre-defined demand parsing model could be an improved LSTM (Long Short-Term Memory) model that has been trained and debugged. Taking the example that the multi-dimensional information includes three dimensions: topic, knowledge attributes, and user characteristics, and the multi-dimensional feature matrix is ​​the aforementioned three-dimensional feature matrix M, the pre-defined demand parsing model could include the following structure: 1. The input layer, for example, can be represented as: X_t=[U_hist,I_real], Wherein, X_t is the input layer, U_hist is the feature data of the user feature dimension in the three-dimensional feature matrix M, and I_real is the real-time interactive behavior data of the target user and the thematic map, such as user browsing map area = "Yunnan Province", user operation = "zoom in map to city level", search keyword = "monsoon influence", etc., and may also include the user's browsing time of the current map area and the user's interaction depth.

[0032] The duration of a user's browsing of the current map area can be a value between 0 and 1, representing the persistence of the user's browsing in the current map area. The closer the duration of browsing of the current map area is to 1, the longer the user's browsing time is, and vice versa. For example, the value can be 0.1 when the user's browsing time is 10 seconds, and 0.5 when the user's browsing time is 30 seconds, etc., but it is not limited to this.

[0033] The aforementioned user interaction depth can be an integer value between 1 and 3 to represent the complexity of the user's operations during the browsing process of the current map area. For example, if the user only browses and then exits during the browsing process of the current map area, the aforementioned user interaction depth can be 1. If the user, in addition to browsing, clicks on each layer to view detailed information during the browsing process of the current map area, the aforementioned user interaction depth can be 2. If the user, in addition to browsing and clicking on each layer to view detailed information during the browsing process of the current map area, also downloads relevant data, the aforementioned user interaction depth can be 3.

[0034] In summary, the real-time interaction data between the target user and the aforementioned thematic map can be, for example, a 5-dimensional vector (i.e., user browsing map area, user operation, search keywords, user browsing time of the current map area, and user interaction depth), but is not limited to this.

[0035] It is understandable that, based on the attention mechanism of the LSTM model, after inputting the above U_hist and I_real into the above X_t, the LSTM model can automatically focus on some dimensions in the above U_hist and I_real data, rather than necessarily using every dimension in the above U_hist and I_real data equally. However, the specific focus can be adjusted and determined according to the actual situation, and no specific restrictions are made here.

[0036] 2. Forget gate, for example, can be represented as: f_t=σ(W_f•[h_{t-1},X_t]+b_f), In this context, f_t is the forget gate, σ() can be a preset function such as the sigmoid function, h_{t-1} is the hidden state at the previous time step, and W_f can be the weight matrix of the forget gate. This weight matrix can be determined by multiplying the hidden layer dimension by (the dimension of the previous hidden layer + the dimension of the input layer), quantifying the influence weights or strength of the hidden state h_{t-1} and the input layer X_t on the forget gate output. Specific values ​​can be adjusted based on actual conditions. It is understood that if the current time step is the first time step (i.e., there is no previous time step), then the hidden state h_{t-1} and the weight matrix W_f of the forget gate can be preset initial values, without specific restrictions here. b_f is the bias term of the forget gate, and its specific value can be adjusted based on actual conditions. It is used to adjust the output baseline of the forget gate, and the dimension of the bias term b_f can be consistent with the hidden layer dimension.

[0037] 3. Input gate and cell state: i_t=σ(W_i·[h_{t-1},X_t]+b_i), c_t=f_t·c_{t-1}+i_t·tanh(W_c·[h_{t-1},X_t]+b_c), In this context, i_t represents the input gate, c_t represents the cell state, W_i is the weight matrix of the input gate (its dimension can be the same as the weight matrix W_f of the forget gate, and its specific value can be adjusted according to the actual situation), and is used to control which information from the input layer X_t and the previous hidden state h_{t-1} is written into the cell state c_t. c_{t-1} represents the cell state at the previous time step, tanh() is the hyperbolic tangent function, W_c is the weight matrix of the cell state (its dimension can be the same as the weight matrix W_f of the forget gate, and its specific value can be adjusted according to the actual situation), and b_c is the bias term of the cell state (its specific value can be adjusted according to the actual situation), used to balance the numerical range of the candidate cell states. The dimension of the cell state bias term b_c can be the same as the dimension of the hidden layer.

[0038] 4. Output gate and real-time demand vector: o_t=σ(W_o·[h_{t-1},X_t]+b_o), h_t = o_t • tanh(c_t), In this context, o_t is the output gate, h_t is the real-time demand vector, W_o is the weight matrix of the output gate (its dimension can be the same as the weight matrix W_f of the forget gate, and its specific value can be adjusted according to the actual situation), and is used to control which information in the cell state c_t is output as the current hidden state h_t, i.e., the real-time demand vector h_t). b_o is the bias term of the output gate (its specific value can be adjusted according to the actual situation), and is used to adjust the activation sensitivity of the output gate. The dimension of the bias term b_o of the output gate can be the same as the dimension of the hidden layer.

[0039] It is understood that the above content is only a possible example. The actual types, structures, and parameters of the preset requirements analysis models can be adjusted and determined according to the actual situation, and are not limited to the examples above.

[0040] S103. Based on the above real-time demand vector and the above multi-dimensional feature matrix, a preset knowledge matching model is used to match each piece of knowledge in the preset knowledge base of the above thematic map to obtain the matching parameters of the above target user for each piece of knowledge.

[0041] For example, based on the aforementioned real-time demand vector and the aforementioned multi-dimensional feature matrix, a preset knowledge matching model is used to match each piece of knowledge in the preset knowledge base of the aforementioned thematic map, thereby obtaining the matching parameters of the target user for each piece of knowledge. These matching parameters can be, for example, a matching score, which can be used to characterize the degree of matching between each piece of knowledge in the preset knowledge base of the thematic map and the aforementioned real-time demand vector h_t. Specifically, taking the aforementioned multi-dimensional information as including three dimensions—topic, knowledge attribute, and user characteristics—and the aforementioned multi-dimensional feature matrix as the aforementioned three-dimensional feature matrix M, the degree of matching between each piece of knowledge in the preset knowledge base of the aforementioned thematic map and the aforementioned real-time demand vector h_t can refer to the degree of matching between the features of the theme dimension and the features of the knowledge attribute dimension corresponding to each piece of knowledge in the preset knowledge base of the aforementioned thematic map in the three-dimensional feature matrix M and the aforementioned real-time demand vector h_t. However, this is not a limitation, and the specific calculation method for the degree of matching between the features of the theme dimension and the features of the knowledge attribute dimension corresponding to each piece of knowledge in the preset knowledge base of the aforementioned thematic map in the three-dimensional feature matrix M and the aforementioned real-time demand vector h_t can be adjusted and determined according to the actual situation, without specific restrictions here.

[0042] S104. Based on the matching parameters of the above knowledge, determine the target knowledge from each of the above knowledge in the preset knowledge base, and push the target knowledge to the target user.

[0043] For example, based on the matching parameters of each of the above-mentioned knowledge items, the target knowledge is determined from each of the above-mentioned knowledge items in the above-mentioned preset knowledge base, and the target knowledge is pushed to the above-mentioned target user. For example, the target knowledge can be determined as the knowledge with the highest matching score, that is, the highest degree of matching with the above-mentioned real-time demand vector h_t (such as 10 items, but not limited to this). Alternatively, after filtering out the preset number of knowledge items with the highest matching score, the target knowledge can be determined in combination with other conditions, and finally the target knowledge can be pushed to the above-mentioned target user. The specific method of determining the target knowledge and the specific content of the above-mentioned other conditions are not limited here, and can be adjusted and determined according to the actual situation.

[0044] The thematic map knowledge intelligent recommendation method provided in this application includes: constructing a multi-dimensional feature matrix of the thematic map based on its multi-dimensional information; dynamically analyzing the target user's needs using a preset demand analysis model based on the real-time interaction data between the target user and the thematic map, and the multi-dimensional feature matrix, to obtain a real-time demand vector of the target user for the thematic map; matching each piece of knowledge in a preset knowledge base of the thematic map using a preset knowledge matching model based on the real-time demand vector and the multi-dimensional feature matrix, to obtain matching parameters for each piece of knowledge for the target user; and determining target knowledge from each piece of knowledge in the preset knowledge base based on the matching parameters, and pushing the target knowledge to the target user. In this embodiment, by collecting multi-dimensional information from the thematic map and establishing a multi-dimensional matrix based on the collected multi-dimensional information to replace the single-dimensional information in the traditional solution, it is possible to dynamically obtain an implementation demand vector representing the user's real-time needs after each real-time user interaction by combining a preset demand analysis model with the multi-dimensional feature matrix. Then, by combining a preset knowledge matching model with the implementation demand vector and the multi-dimensional feature matrix, the matching degree between each piece of information in the preset knowledge base of the thematic map and the user's real-time needs is judged, so as to filter out the knowledge that matches the user's real-time needs and push it, thereby improving the accuracy and diversity of thematic map knowledge recommendation and enhancing the dynamic adaptability of thematic map knowledge recommendation.

[0045] Figure 2 This is a schematic flowchart of a thematic map knowledge-based intelligent recommendation method provided in another embodiment of this application. Optionally, in Figure 1 Based on the embodiments, such as Figure 2 As shown, before constructing the multi-dimensional feature matrix of the thematic map based on its multi-dimensional information, the method may further include: S201. Extract the theme dimension information of the thematic map from the metadata of the thematic map.

[0046] S202. Extract the knowledge attribute dimension information of the above-mentioned thematic maps from the knowledge attribute information corresponding to the above-mentioned thematic maps.

[0047] S203. Extract the user feature dimension information of the above-mentioned thematic map from the user feature information corresponding to the above-mentioned thematic map. The above-mentioned multi-dimensional information includes: the above-mentioned theme dimension information, the above-mentioned knowledge attribute dimension information and the above-mentioned user feature dimension information.

[0048] It is understandable that the above-mentioned multi-dimensional information includes: the above-mentioned topic dimension information, the above-mentioned knowledge attribute dimension information, and the above-mentioned user characteristic dimension information. That is, the dimensions of the above-mentioned multi-dimensional information include three dimensions: topic, knowledge attribute, and user characteristics.

[0049] Furthermore, in the above Figure 2 Based on the embodiments, the aforementioned user characteristic dimension information may include: user knowledge level information, usage scenario information, historical interaction behavior information, and historical feedback information.

[0050] Based on the real-time interaction data between the target user and the thematic map, and the aforementioned multi-dimensional feature matrix, a preset demand analysis model is used to dynamically analyze the target user's demands for the thematic map, resulting in a real-time demand vector for the target user regarding the thematic map. This vector may include: Based on the aforementioned real-time interactive behavior data, and the aforementioned historical interactive behavior information and historical feedback information in the aforementioned user feature dimension information of the aforementioned multi-dimensional feature matrix, the aforementioned demand analysis model is used to dynamically analyze the demands of the aforementioned target users, thereby obtaining the aforementioned real-time demand vector.

[0051] For example, the aforementioned user knowledge level information can be an integer from 1 to 3, and can be based on the user's historical browsing knowledge depth annotation, specifically, it can be beginner=1, intermediate=2, professional=3, but is not limited to this.

[0052] The above-mentioned use case information can also be an integer from 1 to 3, and can be inferred based on the user's selection or historical operations during login. For example, it can be learning=1, research=2, decision support=3, but it is not limited to this.

[0053] The aforementioned historical interaction behavior information may include multi-dimensional information (such as including 3 dimensions, but not limited to this). Specifically, it may include the frequency of clicks on knowledge types and the dwell time of a single knowledge item within a preset historical time period (such as the past 7 days). The dwell time of a single knowledge item may be a score calculated by dividing the dwell time of a single knowledge item (seconds) by 300, but not limited to this.

[0054] The aforementioned historical feedback information may include multi-dimensional information (such as including two dimensions, but not limited to this). Specifically, it may include user ratings of historical recommended knowledge and historical collection actions. The user rating of historical recommended knowledge may be a rating of 0-1, and may be an active evaluation by the user when browsing historical recommended knowledge, such as useful = 1, average = 0.5, irrelevant = 0, etc., but not limited to this. Historical collection actions may also be a rating of 0-1, determined by whether the user has collected each historical recommended knowledge, such as collected = 1, not collected = 0, but not limited to this either.

[0055] Assuming that the total number of features in the above-mentioned topic dimension is 25, the total number of features in the above-mentioned knowledge attribute dimension is 7, and the total number of features in the user feature dimension is 7, then the 25 features of the above-mentioned topic dimension information may include, for example, 5 dimensions of core topic, 12 dimensions of sub-topics, and 8 dimensions of spatial range.

[0056] The seven dimensions of knowledge attribute information mentioned above may include, for example, four dimensions of knowledge type, one dimension of credibility, one dimension of timeliness, and one dimension of relevance.

[0057] The seven dimensions of user feature information mentioned above may include, for example, user knowledge level (1 dimension), usage scenario (1 dimension), historical interaction behavior (3 dimensions), and historical feedback (1 dimension), which correspond to the aforementioned user knowledge level information, usage scenario information, historical interaction behavior information, and historical feedback information, respectively.

[0058] Of course, the above is just one possible example. The total number of features and specific content of the actual topic dimension, knowledge attribute dimension, and user feature dimension may be the same as or different from the above example.

[0059] Continuing with the example above, the matrix elements of the multidimensional feature matrix can be represented as: M[t,a,u], which is the association strength between the t-th topic feature, the a-th knowledge attribute feature, and the u-th user feature. It can be a value between 0 and 1, with higher association strength closer to 1 and lower association strength further away. For example, M["Natural Resources = Mineral Resources", "Credibility = Authoritative Institution", "Knowledge Level = Professional"] = 0.8 can indicate that the association strength between "professional users" and "authoritative institution releases" under the topic of "Natural Resources - Mineral Resources" is relatively high.

[0060] The determination of M[t,a,u] can be achieved, for example, by calculating the interaction frequency of "topic-knowledge-user" in historical interaction behavior information. Specifically, M[t,a,u] = interaction frequency (t,a,u) / total interaction frequency (t,a) to ensure that the initial value of M[t,a,u] conforms to historical patterns. However, the specific determination method of M[t,a,u] is not limited to the above content.

[0061] Based on the above, and using the real-time interactive behavior data, the historical interactive behavior information and historical feedback information in the user feature dimension information of the multi-dimensional feature matrix, the above-mentioned demand analysis model is used to dynamically analyze the demands of the target user to obtain the real-time demand vector. For example, this could mean updating the real-time interactive behavior data every time a user interacts with the thematic map in real time, and using the latest real-time interactive behavior data and the historical interactive behavior information and historical feedback information in the user feature dimension information of the multi-dimensional feature matrix, the above-mentioned demand analysis model is used to dynamically analyze the demands of the target user to obtain the real-time demand vector.

[0062] The aforementioned demand analysis model can dynamically analyze the target user's demands based on the historical interaction behavior information and historical feedback information in the user feature dimension information of the multidimensional feature matrix. For example, it can be based on the attention mechanism of the LSTM model. After inputting the user feature dimension information and real-time interaction behavior data into the aforementioned X_t, the LSTM model can automatically focus on the historical interaction behavior information and historical feedback information in the aforementioned user feature dimension information, rather than necessarily using each dimension in the aforementioned user feature dimension information and real-time interaction behavior data equally.

[0063] Furthermore, based on the above embodiments, before dynamically analyzing the target user's needs using the above-mentioned demand analysis model to obtain the above-mentioned real-time demand vector, according to the above-mentioned real-time interaction behavior data and the above-mentioned historical interaction behavior information and the above-mentioned historical feedback information in the user feature dimension information of the above-mentioned multi-dimensional feature matrix, the above-mentioned demand analysis model may further include: Obtain user thematic map interaction samples, which include: real-time interaction behavior data of the samples, corresponding historical interaction behavior information and historical feedback information of the samples, as well as pre-labeled sample real needs.

[0064] The above-mentioned demand analysis model was obtained by training the model based on the user thematic map interaction samples.

[0065] For example, the above-mentioned sample real-time interaction behavior data, corresponding sample historical interaction behavior information and historical feedback information, as well as user thematic map interaction samples with pre-annotated sample real needs can be used to enable the demand analysis model to understand the logic of "what the target user's real needs are in the context of a certain real-time interaction behavior and corresponding sample historical interaction behavior" from the data and algorithm level.

[0066] In the actual model training process, the above-mentioned user thematic map interaction samples can be divided into training and validation sets. For example, 80% of the sample data in the user thematic map interaction samples can be used as the training set and 20% of the sample data can be used as the validation set. After training on the training set, the real-time interaction behavior data of the samples in the validation set, the corresponding historical interaction behavior information and historical feedback information are used as input values ​​to the trained model. Then, the output of the model is compared with the real needs of the pre-labeled samples in the validation set to verify the loss value of the model.

[0067] The loss value for validating the model can be achieved, for example, through the cross-entropy loss function, but it is not limited to this. If it is achieved through the cross-entropy loss function, the cross-entropy loss function can be in the following form: L=-Σ(y_t·log(h_t)+(1-y_t)·log(1-h_t)), Where L is the loss value of the above model, h_t is the output result of the model, that is, the real-time demand vector output by the model, and y_t is the pre-labeled sample real demand of the validation set, which can be a 0-1 vector, where 0 can represent non-user demand, 1 can represent user demand, but not limited to this.

[0068] Then, the real-time demand vector prediction accuracy of the current model is determined by the loss value L of the above model. When the real-time demand vector prediction accuracy is greater than the preset accuracy threshold, such as greater than 90%, there is no need to adjust the optimization model. Otherwise, the target optimization model can be minimized by an optimizer with a preset initial learning rate to minimize the loss value L of the above model. Specifically, it can be the target optimization model that minimizes the loss value L of the above model by an Adam optimizer with an initial learning rate η=0.01.

[0069] Taking the above-mentioned demand parsing model as an LSTM model as an example, if the real-time demand vector prediction accuracy of the current model is less than the preset minimum accuracy threshold, such as <85%, the weight of "spatial range" in W_f can be increased, for example by increasing it by 0.1, to improve the real-time demand vector prediction accuracy of the demand parsing model.

[0070] Of course, the above is just one possible example, and the actual model evaluation and optimization methods are not limited to the examples above.

[0071] In addition, in the aforementioned Figure 2 Based on the embodiments, the above knowledge matching model includes: an attention weight module and a knowledge matching module. According to the above real-time demand vector and the above multi-dimensional feature matrix, the preset knowledge matching model is used to match each piece of knowledge in the preset knowledge base of the above thematic map, obtaining the matching parameters for each piece of knowledge for the target user, including: Using the aforementioned attention weight module, dynamic weights are calculated for the topic dimension information and the knowledge attribute dimension information in the aforementioned multidimensional feature matrix based on the aforementioned real-time demand vector, respectively, to obtain the first attention weight vector for the topic dimension and the second attention weight vector for the knowledge attribute dimension.

[0072] Using the knowledge matching module described above, the corresponding topic feature vector and knowledge attribute feature vector are extracted from each of the above-mentioned knowledge items in the preset knowledge base. Based on the first attention weight vector, the second attention weight vector, the topic feature vector, and the knowledge attribute feature vector, the matching parameters for each of the above-mentioned knowledge items are determined respectively.

[0073] For example, the attention weight module described above can be represented by the following formula: α_T=softmax(W_T·h_t+b_T), α_A=softmax(W_A·h_t+b_A), Wherein, α_T is the first attention weight vector of the topic dimension, α_A is the second attention weight vector of the knowledge attribute dimension, h_t is the real-time demand vector, b_T and b_A are the bias terms of the topic dimension information attention mechanism and the knowledge attribute dimension information attention mechanism, respectively, and their determination methods and dimensions are similar to b_f, b_i, b_o in the previous embodiments, and will not be repeated here. W_T and W_A are the weight matrices of the topic dimension information and the knowledge attribute dimension information, respectively, and their determination methods and dimensions are similar to W_f, W_i, W_o in the previous embodiments, and will not be repeated here.

[0074] The knowledge matching module mentioned above can be represented by, for example, the following formula: S_k=α_T·T_k^T+α_A·A_k^T+λ·(T_k·M·A_k^T), Here, S_k can be a matching parameter for each piece of knowledge, such as a matching score of 0-100. A score closer to 100 indicates a higher degree of matching between the corresponding knowledge and the real-time demand vector, i.e., the target user's current real-time demand. T_k is the topic feature vector corresponding to the topic dimension information of each piece of knowledge in the multi-dimensional feature matrix. A_k is the knowledge attribute feature vector corresponding to the knowledge attribute features of each piece of knowledge in the multi-dimensional feature matrix. M is the multi-dimensional feature matrix. λ is the correlation coefficient, which can be obtained through the adjustment and optimization process of the demand analysis model. For example, it can be 0.3, but is not limited to this. ^T is not a power of T and is not related to the topic dimension information or the topic feature vector (i.e., not the T in α_T or T_k). It is a transpose symbol used to transpose a row vector into a column vector to suit the corresponding calculations of the knowledge matching module.

[0075] Figure 3 This is a schematic flowchart of a thematic map knowledge-based intelligent recommendation method provided in another embodiment of this application. Optionally, in the foregoing Figure 1 Based on the embodiments, such as Figure 3 As shown, determining the target knowledge from each of the aforementioned knowledge items in the preset knowledge base based on the matching parameters of each of the aforementioned knowledge items can include: S301. Based on the matching parameters of the above knowledge, sort the above knowledge in descending order and select the first preset number of knowledge as multiple candidate knowledge.

[0076] For example, based on the matching parameters of each piece of knowledge, the above-mentioned knowledge is sorted in descending order, and the top preset number of knowledge is selected as multiple candidate knowledge. That is, the knowledge with the highest matching score, that is, the knowledge with the highest degree of matching with the real-time demand vector h_t, is selected as candidate knowledge.

[0077] S302. Based on the preset knowledge constraints, optimize the above multiple candidate knowledge points to obtain optimized knowledge.

[0078] The above-mentioned candidate knowledge is optimized based on preset knowledge constraints to obtain optimized knowledge. For example, optimization can refer to optimizing the candidate knowledge from the perspectives of knowledge type distribution and knowledge relevance, so as to prevent the target knowledge from being too highly related or too concentrated in one category. However, the specific preset knowledge constraints are not limited to the examples above, but can be selected and determined according to actual needs, and are not restricted here.

[0079] S303. Obtain the target knowledge from the optimized knowledge.

[0080] Furthermore, in the above Figure 3 Based on the embodiments, the above-mentioned preset knowledge constraints may include: type constraints.

[0081] Based on preset knowledge constraints, the above-mentioned candidate knowledge is optimized to obtain optimized knowledge, including: Based on the above type constraints, the knowledge type distribution of the above multiple candidate knowledge is statistically analyzed.

[0082] Based on the above knowledge type distribution, determine whether there is a target knowledge type among the multiple candidate knowledge items whose proportion exceeds the preset proportion threshold.

[0083] If the target knowledge type exists among the multiple candidate knowledge items, then the knowledge with the lowest matching parameter among the target knowledge types is selected, and the unselected knowledge with the highest matching parameter among other knowledge types is added to obtain the optimized knowledge.

[0084] For example, assuming there are 30 candidate knowledge items as described above, and the preset percentage threshold is 40%, when the percentage of a certain type of knowledge is greater than 40%, the knowledge items with the lowest matching scores in that type can be removed one by one, and the other types of knowledge items with the highest matching scores can be sequentially added from the results of sorting each knowledge item in descending order, until the optimized knowledge includes the preset number of types, such as at least 3 types of knowledge, but is not limited to the above content.

[0085] In addition, as mentioned above Figure 3 Based on the embodiments, the above-mentioned knowledge constraints may also include: correlation constraints.

[0086] The above-mentioned optimization of multiple candidate knowledge points based on preset knowledge constraints, resulting in optimized knowledge, also includes: Based on the above correlation constraints, calculate the correlation between any two pieces of knowledge among the multiple candidate knowledge items.

[0087] If at least two of the candidate knowledge items have a correlation degree exceeding the preset correlation degree threshold, the knowledge item with the lowest matching parameter among the above at least two related knowledge items is removed to obtain the above optimized knowledge.

[0088] For example, the correlation degree between any two pieces of knowledge among the above multiple candidate knowledge can be calculated by the formula d_k1k2=A_k1•A_k2^T, where d_k1k2 is the correlation degree between the two pieces of knowledge, and A_k1 and A_k2 are the knowledge attribute feature vectors corresponding to the knowledge attribute features of the first and second pieces of knowledge in the multidimensional feature matrix, respectively.

[0089] After completing the thematic map knowledge intelligent recommendation, the above method can also evaluate the recommendation effect based on user feedback and the target user's historical interaction behavior and feedback information after this recommendation. The recommendation effect evaluation can be achieved from three aspects, such as accuracy, diversity, and feedback value.

[0090] In terms of accuracy, it can be determined based on the proportion of knowledge marked as "useful" by customers in the recommended target knowledge. In terms of diversity, it can be determined based on the number of knowledge types in the recommended target knowledge. In terms of feedback value, it can be calculated based on the formula (number of favorites × 2 + number of clicks > 300 seconds × 1 - number of ignored × 1) / total number of recommended target knowledge, but it is not limited to this.

[0091] After obtaining the recommendation effect evaluation for each thematic map knowledge intelligent recommendation, the model parameters of the demand analysis model can be fine-tuned according to the recommendation effect evaluation at a preset period (e.g., every month) to continuously optimize the recommendation effect of thematic map knowledge intelligent recommendation.

[0092] Specifically, this can be achieved, for example, through the following formula: W_new=W_old-η·▽L(W_old), Wherein, W_new refers to the optimized parameter values ​​of W_f, W_i, W_o in the aforementioned embodiments, while W_old refers to the unoptimized parameter values ​​of W_f, W_i, W_o in the aforementioned embodiments. η is the learning rate of the optimizer, which can be, for example, 0.005, but is not limited thereto. ▽L(W_old) is the gradient of the model's loss value L with respect to the parameter W_old, ensuring that the parameters are updated in the direction of "reducing loss and improving performance metrics".

[0093] In addition, the aforementioned multidimensional feature matrix can also be updated according to a preset period (e.g., every week). The update can be achieved, for example, by the following formula: M_new[t,a,u]=M_old[t,a,u]+β·(V_tau-M_old[t,a,u]), Wherein, M_new[t,a,u] is the updated value of matrix element M[t,a,u] of the multidimensional feature matrix in the aforementioned embodiment, and M_old[t,a,u] is the value of matrix element M[t,a,u] of the multidimensional feature matrix in the aforementioned embodiment before the update. β is the learning rate, used to control the update magnitude; for example, it can be 0.2, but is not limited to this. V_tau is the feedback value corresponding to "topic t - attribute a - user u".

[0094] Figure 4This is a schematic diagram of the structure of a thematic map knowledge intelligent recommendation device provided in an embodiment of this application. This thematic map knowledge intelligent recommendation device can execute the aforementioned thematic map knowledge intelligent recommendation method. The device can be integrated into devices with computing power, such as computers and servers. Figure 2 As shown, the device may include: The construction module 410 is used to construct the multidimensional feature matrix of the thematic map based on the multidimensional information of the thematic map.

[0095] The parsing module 420 is used to perform dynamic demand parsing on the target user based on the real-time interaction behavior data between the target user and the aforementioned thematic map, as well as the aforementioned multi-dimensional feature matrix, using a preset demand parsing model, to obtain the real-time demand vector of the target user for the aforementioned thematic map.

[0096] The matching module 430 is used to match each piece of knowledge in the preset knowledge base of the thematic map according to the real-time demand vector and the multi-dimensional feature matrix, and to obtain the matching parameters of the target user for each piece of knowledge.

[0097] The filtering module 440 is used to determine the target knowledge from each of the above-mentioned knowledge in the preset knowledge base according to the matching parameters of each of the above-mentioned knowledge, and push the target knowledge to the target user.

[0098] The thematic map knowledge intelligent recommendation method provided in this application includes: constructing a multi-dimensional feature matrix of the thematic map based on its multi-dimensional information; dynamically analyzing the target user's needs using a preset demand analysis model based on the real-time interaction data between the target user and the thematic map, and the multi-dimensional feature matrix, to obtain a real-time demand vector of the target user for the thematic map; matching each piece of knowledge in a preset knowledge base of the thematic map using a preset knowledge matching model based on the real-time demand vector and the multi-dimensional feature matrix, to obtain matching parameters for each piece of knowledge for the target user; and determining target knowledge from each piece of knowledge in the preset knowledge base based on the matching parameters, and pushing the target knowledge to the target user. In this embodiment, by collecting multi-dimensional information from the thematic map and establishing a multi-dimensional matrix based on the collected multi-dimensional information to replace the single-dimensional information in the traditional solution, it is possible to dynamically obtain an implementation demand vector representing the user's real-time needs after each real-time user interaction by combining a preset demand analysis model with the multi-dimensional feature matrix. Then, by combining a preset knowledge matching model with the implementation demand vector and the multi-dimensional feature matrix, the matching degree between each piece of information in the preset knowledge base of the thematic map and the user's real-time needs is judged, so as to filter out the knowledge that matches the user's real-time needs and push it, thereby improving the accuracy and diversity of thematic map knowledge recommendation and enhancing the dynamic adaptability of thematic map knowledge recommendation.

[0099] Optionally, the aforementioned thematic map knowledge intelligent recommendation device may further include: an extraction module, used to extract thematic dimension information of the thematic map from the metadata of the thematic map; extract the knowledge attribute dimension information of the thematic map from the knowledge attribute information corresponding to the thematic map; and extract the user feature dimension information of the thematic map from the user feature information corresponding to the thematic map. The multi-dimensional information includes: thematic dimension information, knowledge attribute dimension information, and user feature dimension information.

[0100] Optionally, the aforementioned user characteristic dimension information may include: user knowledge level information, usage scenario information, historical interaction behavior information, and historical feedback information.

[0101] The aforementioned parsing module 420 is specifically used to perform dynamic demand parsing on the target user based on the aforementioned real-time interaction behavior data, the aforementioned historical interaction behavior information and the aforementioned historical feedback information in the aforementioned user feature dimension information in the aforementioned multi-dimensional feature matrix, and to obtain the aforementioned real-time demand vector by using the aforementioned demand parsing model.

[0102] Optionally, the aforementioned thematic map knowledge intelligent recommendation device may further include: an acquisition module, used to acquire user thematic map interaction samples, wherein the aforementioned user thematic map interaction samples include: real-time interaction behavior data of the samples, corresponding historical interaction behavior information and historical feedback information of the samples, and pre-labeled sample real needs. Based on the aforementioned user thematic map interaction samples, a model is trained to obtain the aforementioned need parsing model.

[0103] Optionally, the knowledge matching model described above may include an attention weight module and a knowledge matching module.

[0104] The matching module 430 is specifically used to employ the attention weighting module to calculate dynamic weights for the topic dimension information and the knowledge attribute dimension information in the multi-dimensional feature matrix based on the real-time demand vector, thereby obtaining a first attention weight vector for the topic dimension and a second attention weight vector for the knowledge attribute dimension. The knowledge matching module extracts the corresponding topic feature vectors and knowledge attribute feature vectors from each piece of knowledge in the preset knowledge base, and determines the matching parameters for each piece of knowledge based on the first attention weight vector, the second attention weight vector, the topic feature vector, and the knowledge attribute feature vector.

[0105] Optionally, the filtering module 440 is specifically used to sort the knowledge points in descending order according to the matching parameters of each knowledge point, and select the top preset number of knowledge points as multiple candidate knowledge points. Based on preset knowledge constraints, the multiple candidate knowledge points are optimized to obtain optimized knowledge. The target knowledge is then obtained from the optimized knowledge.

[0106] Optionally, the aforementioned preset knowledge constraints may include: type constraints.

[0107] The aforementioned filtering module 440 is specifically used to statistically analyze the knowledge type distribution of the multiple candidate knowledge items based on the aforementioned type constraints. Based on this distribution, it determines whether any target knowledge type exists among the multiple candidate knowledge items, with a proportion exceeding a preset threshold. If the target knowledge type exists among the multiple candidate knowledge items, the knowledge item with the lowest matching parameter among these target knowledge types is selected, and other unselected knowledge items with the highest matching parameters among the other knowledge types are added to obtain the optimized knowledge.

[0108] Optionally, the aforementioned preset knowledge constraints may also include: correlation constraints.

[0109] The aforementioned filtering module 440 can also be used to calculate the correlation between any two pieces of knowledge among the multiple candidate knowledge items based on the aforementioned correlation constraints. If there are at least two pieces of related knowledge items among the multiple candidate knowledge items whose correlation exceeds a preset correlation threshold, the knowledge item with the lowest matching parameter among the at least two pieces of related knowledge is removed, and the optimized knowledge is obtained.

[0110] The above-described device is used to execute the method provided in the foregoing embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

[0111] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device can be a computer, server, or other device with computing power, as described above. Figure 5 As shown, the device 500 includes: The processor 510, storage medium 520, and bus 550 are connected in communication via bus 550.

[0112] The storage medium 520 stores machine-readable instructions that can be executed by the processor 510. When the electronic device is running, the processor 510 executes the aforementioned machine-readable instructions to perform the aforementioned thematic map knowledge intelligent recommendation method.

[0113] It should be understood that, Figure 5 The structure shown is only a schematic diagram of an electronic device; the electronic device may also include components that are larger than those shown. Figure 5 The more or fewer components shown, or having the same Figure 5 The different configurations shown. Figure 5 The components shown can be implemented using hardware, software, or a combination thereof.

[0114] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the thematic map knowledge intelligent recommendation method described in the above method embodiments.

[0115] Computer-readable storage media can be electronic storage devices such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Optionally, computer-readable storage media include non-transitory computer-readable storage media. The computer-readable storage medium has storage space for program code that performs any of the method steps described above. This program code can be read from or written to one or more computer program products. The program code can be compressed, for example, in a suitable form.

[0116] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0117] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0118] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 in the various embodiments of this application. 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.

[0119] The above description is merely a preferred embodiment of this application and does not limit the patent scope of this application. Any equivalent structural transformations made based on the inventive concept of this application and the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the patent protection scope of this application.

Claims

1. A method for intelligent recommendation of thematic map knowledge, characterized in that, include: Based on the multi-dimensional information of the thematic map, a multi-dimensional feature matrix of the thematic map is constructed; Based on the real-time interaction data between the target user and the thematic map, and the multi-dimensional feature matrix, a preset demand analysis model is used to dynamically analyze the target user's demand for the thematic map, thereby obtaining the target user's real-time demand vector for the thematic map. Based on the real-time demand vector and the multi-dimensional feature matrix, a preset knowledge matching model is used to match each piece of knowledge in the preset knowledge base of the thematic map to obtain the matching parameters of the target user for each piece of knowledge. Based on the matching parameters of each piece of knowledge, target knowledge is determined from each piece of knowledge in the preset knowledge base, and the target knowledge is pushed to the target user; Before constructing the multidimensional feature matrix of the thematic map based on its multidimensional information, the method further includes: Extract the thematic dimension information of the thematic map from its metadata; Extract the knowledge attribute dimension information of the thematic map from the knowledge attribute information corresponding to the thematic map; The user feature dimension information of the thematic map is extracted from the user feature information corresponding to the thematic map. The multi-dimensional information includes: the theme dimension information, the knowledge attribute dimension information, and the user feature dimension information. The knowledge matching model includes an attention weighting module and a knowledge matching module; the step of matching each piece of knowledge in the preset knowledge base of the thematic map using a preset knowledge matching model based on the real-time demand vector and the multi-dimensional feature matrix to obtain the matching parameters of the target user for each piece of knowledge includes: Using the attention weight module, dynamic weights are calculated for the topic dimension information and the knowledge attribute dimension information in the multidimensional feature matrix according to the real-time demand vector, to obtain the first attention weight vector of the topic dimension and the second attention weight vector of the knowledge attribute dimension. The knowledge matching module extracts the corresponding topic feature vector and knowledge attribute feature vector from each piece of knowledge in the preset knowledge base, and determines the matching parameters of each piece of knowledge based on the first attention weight vector, the second attention weight vector, the topic feature vector, and the knowledge attribute feature vector.

2. The method according to claim 1, characterized in that, The user characteristic dimension information includes: user knowledge level information, usage scenario information, historical interaction behavior information, and historical feedback information; The step involves using a preset demand analysis model to dynamically analyze the target user's demands based on real-time interaction data between the target user and the thematic map, and the multi-dimensional feature matrix, to obtain the target user's real-time demand vector for the thematic map, including: Based on the real-time interaction behavior data, and the historical interaction behavior information and historical feedback information in the user feature dimension information of the multi-dimensional feature matrix, the demand parsing model is used to dynamically parse the target user's demand to obtain the real-time demand vector.

3. The method according to claim 2, characterized in that, Before obtaining the real-time demand vector by dynamically analyzing the target user's demands using the demand analysis model based on the real-time interaction behavior data and the historical interaction behavior information and historical feedback information in the user feature dimension information of the multi-dimensional feature matrix, the method further includes: Obtain user thematic map interaction samples, wherein the user thematic map interaction samples include: real-time interaction behavior data of the samples, corresponding historical interaction behavior information and historical feedback information of the samples, and pre-labeled sample real needs; The demand analysis model is obtained by training the model based on the user thematic map interaction samples.

4. The method according to claim 1, characterized in that, The step of determining the target knowledge from each piece of knowledge in the preset knowledge base based on the matching parameters of each piece of knowledge includes: Based on the matching parameters of each piece of knowledge, the pieces of knowledge are sorted in descending order, and the first preset number of pieces of knowledge are selected as multiple candidate pieces of knowledge. Based on preset knowledge constraints, the multiple candidate knowledge items are optimized to obtain optimized knowledge. The target knowledge is obtained from the optimized knowledge.

5. The method according to claim 4, characterized in that, The preset knowledge constraints include: type constraints; The optimization of the multiple candidate knowledge items based on preset knowledge constraints to obtain optimized knowledge includes: Based on the type constraints, statistically analyze the knowledge type distribution of the multiple candidate knowledge items; Based on the knowledge type distribution, determine whether there is a target knowledge type among the multiple candidate knowledge items whose proportion exceeds a preset proportion threshold; If the target knowledge type exists among the multiple candidate knowledge items, then the knowledge item with the lowest matching parameter among the target knowledge types is selected, and the unselected knowledge items with the highest matching parameters among other knowledge types are added to obtain the optimized knowledge.

6. The method according to claim 4, characterized in that, The preset knowledge constraints also include: relevance constraints; The step of optimizing the multiple candidate knowledge items according to preset knowledge constraints to obtain optimized knowledge further includes: Based on the aforementioned correlation constraint, calculate the correlation between any two pieces of knowledge among the multiple candidate knowledge items; If at least two of the candidate knowledge items have a correlation degree exceeding a preset correlation degree threshold, the knowledge item with the lowest matching parameter among the at least two candidate knowledge items is removed to obtain the optimized knowledge.

7. A thematic map knowledge intelligent recommendation device, characterized in that, include: The construction module is used to construct a multi-dimensional feature matrix of the thematic map based on the multi-dimensional information of the thematic map; The parsing module is used to perform dynamic demand parsing on the target user based on the real-time interaction behavior data between the target user and the thematic map, as well as the multi-dimensional feature matrix, using a preset demand parsing model to obtain the real-time demand vector of the target user for the thematic map. The matching module is used to match each piece of knowledge in the preset knowledge base of the thematic map according to the real-time demand vector and the multi-dimensional feature matrix, using a preset knowledge matching model, to obtain the matching parameters of the target user for each piece of knowledge; The filtering module is used to determine target knowledge from each piece of knowledge in the preset knowledge base according to the matching parameters of each piece of knowledge, and push the target knowledge to the target user; The thematic map knowledge intelligent recommendation device further includes: an extraction module, used to extract the theme dimension information of the thematic map from the metadata of the thematic map; extract the knowledge attribute dimension information of the thematic map from the knowledge attribute information corresponding to the thematic map; and extract the user feature dimension information of the thematic map from the user feature information corresponding to the thematic map, wherein the multi-dimensional information includes: the theme dimension information, the knowledge attribute dimension information, and the user feature dimension information; The knowledge matching model includes: an attention weight module and a knowledge matching module; The matching module is specifically used to employ the attention weight module to calculate dynamic weights for the topic dimension information and the knowledge attribute dimension information in the multidimensional feature matrix according to the real-time demand vector, thereby obtaining a first attention weight vector for the topic dimension and a second attention weight vector for the knowledge attribute dimension. The knowledge matching module extracts the corresponding topic feature vector and knowledge attribute feature vector from each piece of knowledge in the preset knowledge base, and determines the matching parameters of each piece of knowledge based on the first attention weight vector, the second attention weight vector, the topic feature vector, and the knowledge attribute feature vector.

8. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of the method as described in any one of claims 1-6.

Images