A science popularization knowledge accurate pushing method and system
By constructing a science knowledge system map and dividing communities, and quantifying users' cognitive status and influence, the accuracy and personalization issues of science popularization platform recommendations are solved, thereby improving the accuracy of recommendations and the learning experience.
Patent Information
- Application Number
- CN202610247922.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-02
- Publication Date
- 2026-06-09
AI Technical Summary
Existing science popularization knowledge recommendation technologies, when applied to science popularization platforms, struggle to accurately depict users' cognitive states and neglect the internal structure of knowledge and the socialized learning environment, resulting in recommendations that are not precise or personalized enough.
By constructing a science knowledge system map, quantifying the characteristic vector of users' cognitive state, dividing dynamic learning communities, and evaluating the influence weight of individuals in the community, a two-level matching calculation is performed by combining individual cognitive level and community environmental factors to achieve accurate science knowledge delivery.
It achieves accurate recommendations that match the user's individual cognitive development stage and reflect the community's learning trajectory, thereby improving the accuracy, personalization, and learning experience of the recommendations.
Smart Images
Figure CN122173707A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of popular science knowledge technology, and in particular to a method and system for accurately delivering popular science knowledge. Background Technology
[0002] With the rapid development of internet technology and the acceleration of educational informatization, a vast amount of popular science content is being widely disseminated through digital platforms. To improve user learning efficiency and experience, personalized recommendation technology has been introduced into the field of popular science. This technology aims to precisely select knowledge resources from the massive amount of content that match users' interests and cognitive levels, thereby alleviating information overload and stimulating and maintaining users' learning motivation.
[0003] Currently, personalized recommendation technologies applied to science popularization platforms are mainly borrowed from e-commerce and news information fields. These technologies are typically based on users' historical behavioral data, analyzing user interactions such as clicks, browsing, and favorites to build user interest models, and then recommending content that users may be interested in. Common methods include collaborative filtering-based recommendations, content tag matching-based recommendations, and sequence recommendations that combine temporal behavior. These methods can, to some extent, improve content exposure and user click-through rates.
[0004] However, the learning process of popular science knowledge has its inherent unique characteristics. Unlike general information consumption, popular science learning is often accompanied by a gradual improvement in cognitive level, and the knowledge content usually has a logical sequence and a systematic structural connection. At the same time, users' learning behavior on the platform is not limited to individual content consumption, but may also include social interactions such as question-and-answer sessions, discussions, and collaborations within the community. These factors mean that when general recommendation technology is directly applied to popular science scenarios, it faces challenges in several aspects, including how to more precisely characterize users' cognitive states, how to effectively utilize the inherent structure of knowledge, and how to integrate it into a social learning environment. Summary of the Invention
[0005] This application provides a method for accurately pushing popular science knowledge. By constructing and quantifying the user's cognitive state feature vector, dividing dynamic learning communities based on cognitive similarity, evaluating the influence weight of individuals in the community, and integrating individual cognitive level and community environmental factors for two-level matching calculation, it ultimately achieves accurate popular science knowledge push that not only conforms to the user's personal cognitive development stage, but also reflects the learning context and trend of their community.
[0006] This application provides a method for accurately delivering popular science knowledge, including: S101, acquire the interactive behavior data of the target users on the science popularization knowledge platform within a preset time period, and form a time-series user behavior sample set; S102, based on the pre-constructed popular science knowledge system map and the user behavior sample set of the target users, use the preset cognitive assessment algorithm to evaluate the cognitive level score of the target users on at least one popular science knowledge dimension defined in the popular science knowledge system map, and generate a cognitive state feature vector. S103: Using a pre-defined clustering algorithm, all active users of the popular science knowledge platform are divided into at least one learning community. Each learning community corresponds to a group cognitive level vector. The knowledge influence weight of the target user in the learning community is calculated. S104. For each candidate popular science knowledge content item, based on its associated popular science knowledge dimensions and content characteristics, combined with the cognitive state feature vector and knowledge influence weight of the target user, a comprehensive matching degree is obtained. S105: Sort all candidate science popularization content items in descending order according to the overall matching degree, generate a target push list, and push it to the target user terminal.
[0007] Preferably, the method for constructing the popular science knowledge system map is as follows: We acquire popular science content from popular science knowledge platforms, extract the popular science knowledge dimension and difficulty level value tags for each popular science knowledge content item, and construct a triplet knowledge graph with the structure of "popular science knowledge dimension - connection edge - content item" based on the connection relationship between popular science knowledge dimensions.
[0008] Preferably, S102 specifically includes: Based on the user behavior sample set, obtain the interaction behavior data of the target users on each predefined dimension of popular science knowledge in the popular science knowledge system map; Based on content interaction metrics from interactive behavior data, a pre-defined cognitive state assessment model is used to calculate the cognitive level score of the target user in at least one predefined science knowledge dimension. This process iterates through the science knowledge dimensions in the science knowledge system graph to generate the user's cognitive state feature vector. , Let m be the cognitive level score for the i-th knowledge dimension, and m be the total number of predefined knowledge dimensions in the graph. .
[0009] Preferably, the preset cognitive assessment algorithm is configured to perform a weighted fusion of the average reading time, average answer accuracy, and content review rate under the corresponding knowledge dimension: The average reading time, average quiz accuracy, and content revisit rate were all obtained from a user behavior sample set. This represents the average reading time of the target user within knowledge dimension i. The average correct answer rate of the target users in knowledge dimension i. This indicates the user's content replay rate under knowledge dimension i. For the weight parameters, satisfying , This is the normalized benchmark value for the corresponding indicator, used for normalization processing.
[0010] Preferably, the step of using a preset clustering algorithm to divide all active users of the science popularization platform into at least one learning community and its group cognitive level vector specifically includes: Based on the cognitive state feature vectors of all active users, a pre-defined clustering algorithm is used to cluster them to generate at least one learning community. Each learning community includes at least one active user and its cognitive state feature vector.
[0011] Preferably, the group cognitive level vector of the learning community is set as follows: the average cognitive level score of each active user on each knowledge dimension is calculated, that is, the average feature vector of the cognitive state feature vector of all active users, forming a group cognitive level vector representing the overall knowledge level of the community. .
[0012] Preferably, the step of combining the target user's cognitive state feature vector and their knowledge influence weight within their learning community to obtain a comprehensive matching degree specifically includes: Calculate the first matching score between each candidate science popularization knowledge content item and the cognitive state feature vector of the target user, and the second matching score between each candidate science popularization knowledge content item and the knowledge influence weight of the target user's community. Then, weight and fuse the first matching score and the second matching score to obtain the comprehensive matching degree.
[0013] Preferably, the second matching score is calculated according to the second matching algorithm. : in, The knowledge influence factor of target user a within their community g. Let k be the set of science knowledge dimensions associated with the candidate science knowledge content item k. This represents the number of science knowledge dimensions associated with candidate science knowledge content item k. The average cognitive level score of community g on the knowledge dimension c. This represents the learning popularity of content within the knowledge dimension c in community g within a preset time window. and These are the preset balance coefficients. This is used to adjust the relative importance of community level fit and community learning enthusiasm.
[0014] Preferably, S105 specifically includes: The overall matching degree of all candidate popular science knowledge content items is sorted in descending order to generate an initial recommendation sequence. The top three content items in the initial recommendation sequence are selected to form the target push list.
[0015] This application also provides a system for accurately pushing popular science knowledge, including: an acquisition module, an evaluation module, a matching module, and a generation module; The acquisition module is used to acquire interactive behavior data of target users within a preset time period on the popular science knowledge platform, forming a time-series user behavior sample set. The evaluation module is used to evaluate the cognitive level score of the target user in at least one science knowledge dimension defined in the science knowledge system graph based on the pre-constructed science knowledge system graph and the user behavior sample set of the target user, and to generate a cognitive state feature vector. It also uses a pre-constructed clustering algorithm to divide all active users of the science platform into at least one learning community and its group cognitive level vector, and calculates the knowledge influence weight of the target user in its learning community. The matching module is used to obtain a comprehensive matching degree for each candidate popular science knowledge content item based on its associated popular science knowledge dimensions and content characteristics, combined with the cognitive state feature vector and knowledge influence weight of the target user. The generation module is used to sort all candidate popular science knowledge content items in descending order according to the comprehensive matching degree, generate a target push list, and push it to the target user terminal.
[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: By constructing a structured science knowledge system graph, fine-grained learning behavior data of users is mapped into cognitive state feature vectors, enabling precise quantification and dynamic evaluation of individual knowledge mastery. Furthermore, users are dynamically divided into communities based on cognitive characteristics, and the influence weight of members is quantified based on the knowledge dissemination network within the community, forming a collaborative representation of individual cognitive profiles and the group learning environment. A two-layer matching strategy integrating personal cognitive fit with community cognitive trends and popularity is employed to generate recommendations, ensuring that the pushed content accurately matches the user's current cognitive level and cognitive development path, while also reflecting the collective wisdom and learning trends of their learning community. Finally, closed-loop optimization of the cognitive model and recommendation strategy is achieved through user feedback, thus solving key problems in science popularization recommendations such as coarse cognitive assessment, fragmented knowledge systems, lack of social learning factors, and cold start, significantly improving the accuracy, personalization, rationality of learning paths, and user learning experience and sustainability. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating the method for accurately pushing popular science knowledge according to an embodiment of the present invention. Figure 2 This is a structural block diagram of the science popularization knowledge accurate push system according to an embodiment of the present invention. Detailed Implementation
[0018] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.
[0019] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.
[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0021] Example 1: Figure 1 This is a flowchart illustrating the method for accurately pushing popular science knowledge according to an embodiment of the present invention.
[0022] like Figure 1 As shown, a method for accurately delivering popular science knowledge includes the following steps: S101, acquire the interactive behavior data of the target user within a preset time period on the popular science knowledge platform, organize the behavior data in chronological order, and store it with user identifier and content identifier as a joint index to form a time-seriesd user behavior sample set.
[0023] Among them, the preset time period Configure the settings according to the actual push scenario; for example, set it to the past week.
[0024] For example, all user interaction events (i.e., interaction behavior data) on the science popularization knowledge platform are recorded in real time, including but not limited to: content clicks, content learning completion markers, participation in embedded knowledge tests and correct / incorrect results, page dwell time, proactive questioning, collection and sharing, etc.; each interaction event includes a timestamp, user ID, content ID, event type and event-related data (such as test score, dwell time, etc.); all interaction events are arranged in ascending order by timestamp to construct a time-series behavior chain for each user; using (user ID, content ID) as a joint primary key and binding with related science popularization knowledge dimension tags, the interaction behavior data is stored in a time-series database to form a structured user behavior sample set, supporting efficient querying and aggregation by user, content, and time window.
[0025] Furthermore, the interactive behavior data may also include content interaction metrics and community interaction metrics. Content interaction metrics include, but are not limited to, the target user's reading time, quiz accuracy, number of times content is reviewed, frequency of content tagging, and content collection behavior across various science popularization knowledge dimensions. Community interaction metrics include, but are not limited to, the number of times the target user asks questions, the number of times their answers are accepted (which can be defined as likes, comments, and other interactive behaviors), the number of times content is shared, and the frequency of participation in co-learning groups. Community interaction metrics can also be classified into explicit social interaction and implicit social interaction. For example, explicit social interaction includes following, liking, commenting, and mentioning, while implicit social interaction includes content sharing paths and co-learning records.
[0026] S102, based on the pre-constructed popular science knowledge system map and the user behavior sample set of the target user, the cognitive level score of the target user on at least one popular science knowledge dimension defined in the popular science knowledge system map is evaluated using a preset cognitive evaluation algorithm, and a cognitive state feature vector is generated.
[0027] For example, a science knowledge system map can be: Acquire science popularization content data from a science popularization knowledge platform. The science popularization content data includes multimodal content items. Semantic parsing is performed on each science popularization content item to extract its core knowledge unit. Each core knowledge unit is defined as a science popularization knowledge dimension, such as "Newton's First Law", "the light reaction stage of photosynthesis", and "the consensus mechanism of blockchain". Through semantic normalization processing, science popularization knowledge dimensions with different expressions but the same meaning are merged to form a standardized set of science popularization knowledge dimensions. Define and construct the connection relationship between any two knowledge dimensions in the set of popular science knowledge dimensions. The connection relationship includes at least the prerequisite relationship, logical relationship and cross-domain relationship. Assign a relationship weight to each connection relationship based on domain experts. The relationship weight is used to characterize the association strength or learning dependence. With the popular science knowledge dimension as the node and the connection relationship as the edge, the initial popular science knowledge graph network structure is constructed. Standardized difficulty levels are assigned to each dimension of popular science knowledge. The difficulty level values are generated based on settings set by domain experts (normalized to the range of 0 to 1), and a corresponding multi-dimensional feature vector is generated for each dimension of popular science knowledge. The multidimensional feature vector is obtained by fusing the semantic embedding features of the dimension, the network structure of the relationship, and the statistical features of the associated content, and is used to uniquely represent the semantics and attributes of the dimension in the vector space. Each science popularization content item is associated with one or more science popularization knowledge dimensions. Using the extracted science popularization knowledge dimensions, it is then mapped to one or more corresponding dimension nodes in the graph, establishing a mapping relationship between content items and dimension nodes. This generates a content feature vector for each science popularization content item. And ensure that the content feature vector and its associated multidimensional feature vector of popular science knowledge are in the same or mappable vector space where similarity calculation can be performed.
[0028] It should be noted that in the connection relationships, prerequisite relationships indicate that learning dimension A requires prior knowledge of dimension B, such as "calculus" being prerequisite knowledge for "Newtonian mechanics." This relationship originates from subject syllabi, textbook catalogs, and expert knowledge. Logical association relationships indicate that dimensions are logically closely related, such as "causal relationship" (greenhouse gas emissions → global warming), "compositional relationship" (the cell nucleus is a component of the cell), and "analogical relationship" (electric current is analogous to water flow). This relationship is obtained through co-occurrence analysis, semantic similarity calculation (such as sentence vector similarity based on BERT), and rule mining. Cross-domain association relationships reveal implicit connections between dimensions of different disciplines to support interdisciplinary learning paths, such as "fractal geometry" and "coastline morphology," and "quantum entanglement" and "holism in philosophy." This relationship relies on cross-textual citation analysis under big data, concept co-occurrence networks, and expert definitions. This invention will not elaborate on these aspects.
[0029] Specifically, the method for constructing a science knowledge system map is as follows: We acquire popular science content from popular science knowledge platforms, extract the popular science knowledge dimension and difficulty level value tags for each popular science knowledge content item, and construct a triplet knowledge graph with the structure of "popular science knowledge dimension - connection edge - content item" based on the logical relationship, prerequisite relationship and cross-domain relationship between popular science knowledge dimensions. We also assign a multi-dimensional feature vector to each popular science knowledge dimension and generate a content feature vector for the knowledge content item that is semantically aligned with the multi-dimensional feature vector of the popular science knowledge dimension.
[0030] It should be noted that each science popularization knowledge item is labeled with one or more associated knowledge dimensions and its contribution strength to those knowledge dimensions when it is entered into the database.
[0031] In some embodiments, step S102 specifically includes: Based on the user behavior sample set, obtain the interaction behavior data of the target users on the content of each predefined science knowledge dimension in the science knowledge system map; Based on content interaction metrics from interactive behavior data, a pre-defined cognitive state assessment model is used to calculate the cognitive level score of the target user in at least one predefined science knowledge dimension. This process iterates through the science knowledge dimensions in the science knowledge system graph to generate the user's cognitive state feature vector. , Let m be the cognitive level score for the i-th knowledge dimension, and m be the total number of predefined knowledge dimensions in the graph. .
[0032] In some embodiments, the cognitive state feature vector is constructed based on content interaction metrics: Using a pre-defined cognitive assessment algorithm, the cognitive level score of the target user in at least one predefined dimension of popular science knowledge is calculated. These constitute the cognitive state feature vector.
[0033] Specifically, the preset cognitive assessment algorithm is set to weight and fuse the average reading time, average answer accuracy, and content review rate for the corresponding knowledge dimension: The average reading time, average quiz accuracy, and content revisit rate were all obtained from a user behavior sample set. This represents the average reading time of the target user under knowledge dimension i (based on the average reading time of each reading session). The average correct answer rate of the target users in knowledge dimension i. This represents the user's content replay rate under knowledge dimension i, which is the proportion of replayed content out of the total number of accessed content. For the weight parameters, satisfying The importance is preset based on reading time, quiz accuracy, and number of replays. This is the normalized baseline value for the corresponding metric, set to the maximum value for all users on the corresponding metric, used for normalization processing.
[0034] S103. Based on the cognitive state feature vector of the target user, use a preset clustering algorithm to divide all active users of the popular science knowledge platform (i.e., users who have interactive behavior data on the popular science knowledge platform within a preset time period) into at least one learning community. Each learning community corresponds to a group cognitive level vector. Based on the user behavior sample set associated with the learning community to which the target user belongs, calculate the knowledge influence weight of the target user in the learning community to which the target user belongs.
[0035] Specifically, the preset clustering algorithm can be set to density-based DBSCAN, which uses the cognitive state feature vectors of all active users as the core features to generate at least one learning community, with each learning community including at least one active user.
[0036] For each learning community g, its group cognitive level vector is set as follows: the average cognitive level score of each active user on each knowledge dimension is calculated, which is the average feature vector of the cognitive state feature vector of all active users, forming the group cognitive level vector representing the overall knowledge level of the community. .
[0037] In some embodiments, calculating the knowledge influence weight of a target user within a community based on a user behavior sample set associated with the community specifically includes: S201, based on each learning community, constructs a directed graph of knowledge propagation within the community using community interaction metrics from a set of user behavior samples of all active users. Nodes represent users, and edge weights... This is determined based on information about social interactions between users.
[0038] Specifically, based on community interaction metrics, social interaction information among active users within the community is obtained. This information includes the frequency of social interactions, the type of each interaction, and a predefined interaction weight value for each type. For example, social interaction types may include: likes, comments, following, collaborative learning, and sharing.
[0039] Among them, the weight of the edge This represents the strength of user u's influence on user v. It is calculated by averaging the interaction weights of each social interaction between user u and user v.
[0040] S202, using a pre-defined influence propagation model, calculate the knowledge influence weight of target user a within their community g. .
[0041] Specifically, the preset influence propagation model is set as follows: in, Let represent the set of all users who have influenced target user 'a' (users who have engaged in social interactions with user 'u') in the directed graph of knowledge dissemination. This represents the set of users that user v has influenced. This represents the strength of user v's influence on user a, and can be obtained by weighted summation based on interaction frequency and type. The weight of interaction type can be set by domain experts. This is a preset damping factor, typically set to 0.85, representing the attenuation rate of influence propagation along the network. The higher the value, the more important user a's position as a knowledge contributor or learning leader in community g. The underlying principle can be found in descriptions of the PageRank algorithm in related technologies; this invention will not elaborate on that.
[0042] Optionally, in other embodiments, the knowledge influence weight of a target user within their community can also be calculated by weighting and integrating the target user's answer adoption rate, content sharing and dissemination rate, and number of group learning sessions under the corresponding knowledge dimension i within the community, thus calculating the target user's knowledge influence factor within their community. : in, The adoption rate of the answers given by target user A within the community. The content sharing and dissemination rate of target user a within the community can be defined as the probability that all content shared by target user a within the community will be disseminated secondaryly. The number of times target user A initiates or participates in group learning within the community. , , These are the normalized baseline values for the corresponding metrics, set to the maximum value for the corresponding metrics across all active users within the community, used for normalization. To be respectively the corresponding preset weight coefficients, satisfying The settings are pre-defined based on the importance of the corresponding indicators.
[0043] S104. For each candidate science popularization knowledge content item, based on its associated science popularization knowledge dimensions and content characteristics, combined with the target user's cognitive state feature vector and its knowledge influence weight within its learning community, a comprehensive matching degree is obtained.
[0044] In some embodiments, a comprehensive matching degree is obtained by combining the cognitive state feature vector of the target user and the knowledge influence weight within their learning community, specifically including: The system calculates the first matching score between each candidate science knowledge content item and the target user's cognitive state feature vector, and the second matching score between each candidate item and the target user's community's knowledge influence weight. The first and second matching scores are then weighted and fused to obtain the overall matching degree. The fusion weights of the first and second matching scores are set according to the actual situation, with a sum of 1 being sufficient. For example, the initial value is biased towards the individual user, with the first matching score fusion weight set to 0.7 and the second matching score fusion weight set to 0.3. The system dynamically adjusts these weights based on the user's feedback on the differences between individual-oriented and community-oriented content in historical recommendations.
[0045] Specifically, for a candidate science knowledge content item k, let its associated set of science knowledge dimensions be . The first matching score is calculated based on the first matching algorithm: in, This indicates the number of science knowledge dimensions associated with content item k. For the feature vector of dimension c in the pre-constructed popular science knowledge system graph, represent its semantics and attributes. This represents the content feature vector of content item k. This represents the cosine similarity function, used to measure the semantic relevance between content and knowledge dimensions. For the pre-defined standardized difficulty level value of knowledge dimension c, map to interval, Rate the cognitive level of target user a in knowledge dimension c. This indicates the gap between the difficulty of the content in the knowledge dimension c and the current cognitive level of the target user a. The smaller the gap, the higher the suitability.
[0046] The second matching score is calculated based on the second matching algorithm: in, The knowledge influence factor of target user a within their community g. The average cognitive level score of community g on the knowledge dimension c. This represents the learning popularity of content within the knowledge dimension c in community g within a preset time window. , This represents the number of interaction events in community g within a preset time window, under the knowledge dimension c. This represents the maximum number of interaction events for community g across all knowledge dimensions within a preset time window (less than the preset time period length). and These are the preset balance coefficients. This is used to adjust the relative importance of community level fit and community learning enthusiasm.
[0047] S105: Sort all candidate science popularization content items in descending order according to the overall matching degree, generate a target push list, and push it to the target user terminal.
[0048] Specifically, the comprehensive matching degree of all candidate popular science knowledge content items is sorted in descending order to generate an initial recommendation sequence. The top three content items are selected from the initial recommendation sequence, and logically rearranged according to the connection relationship (prerequisite and association relationship) between knowledge dimensions in the pre-constructed popular science knowledge system map to form a target push list that conforms to the cognitive development path, which is then delivered to the target user terminal through the push channel.
[0049] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: The system acquires interactive behavior data of target users within a preset time period on a science popularization knowledge platform, forming a time-series user behavior sample set. Based on a pre-constructed science popularization knowledge system graph and the target user's user behavior sample set, a preset cognitive assessment algorithm is used to evaluate the target user's cognitive level score on at least one science popularization knowledge dimension defined in the science popularization knowledge system graph, generating a cognitive state feature vector. According to the target user's cognitive state feature vector, a preset clustering algorithm is used to divide all active users on the science popularization platform into at least one learning community and its group cognitive level vector. Based on the user behavior sample set associated with the target user's learning community, the system calculates the target user's knowledge influence weight within the learning community. For each candidate science popularization knowledge content item, a comprehensive matching degree is obtained based on its associated science popularization knowledge dimension and content characteristics, combined with the target user's cognitive state feature vector and its knowledge influence weight within the learning community. All candidate science popularization knowledge content items are sorted in descending order according to the comprehensive matching degree to generate a target push list, which is then pushed to the target user.
[0050] It achieves refined modeling and dynamic evaluation of users' cognitive states. By mapping users' fine-grained behavioral indicators across multiple dimensions onto a structured science popularization knowledge system, a quantified cognitive state feature vector is generated. This vector can more accurately and comprehensively reflect users' current cognitive level and knowledge mastery, laying a reliable data foundation for subsequent precise matching. A learning community segmentation mechanism integrating knowledge structure and cognitive characteristics was constructed. Clustering was performed based on users' cognitive state feature vectors, dynamically aggregating users with similar cognitive patterns into learning communities, and calculating the group cognitive level vector of each community. This approach not only considers the similarity of users' interests but also emphasizes the convergence of their cognitive development stages, thus enabling more effective identification of user groups with similar learning needs and paths. The knowledge influence of users in learning communities was quantitatively assessed. By analyzing users' social interaction behaviors within the community (such as Q&A, sharing, and collaborative learning), a knowledge dissemination network was constructed and its influence weight was calculated. This enabled the identification of knowledge contributors and key nodes in the dissemination within the community, providing a quantitative basis for integrating social learning factors into recommendations. A two-tiered matching recommendation strategy integrating individual cognitive level and community environment factors is proposed. When calculating content matching degree, both the fit between the content and the user's individual cognitive state and the fit between the content and the overall cognitive level and learning enthusiasm of the user's community are considered. This is further adjusted by the user's personal influence weight, ensuring that the recommendation results not only match the individual's current learning stage but also reflect the learning atmosphere and trends of their community, thus enhancing the personalization and contextual adaptability of the recommendations.
[0051] Therefore, by comprehensively assessing the individual cognitive state of users and quantifying the learning environment of their respective communities, the system achieves the integration of individual needs and group trends in the delivery of popular science knowledge. Compared with traditional methods, this approach can more accurately adapt to the cognitive development patterns of users and meet their social learning needs.
[0052] Furthermore, this embodiment of the invention also provides a system for precisely delivering popular science knowledge.
[0053] Figure 2 This is a structural block diagram of a science popularization knowledge accurate push system according to an embodiment of the present invention.
[0054] like Figure 2 As shown, a precise science popularization knowledge delivery system includes: an acquisition module, an evaluation module, a matching module, and a generation module; The acquisition module is used to acquire interactive behavior data of target users within a preset time period on the popular science knowledge platform, forming a time-series user behavior sample set. The evaluation module is used to evaluate the cognitive level score of the target user in at least one science knowledge dimension defined in the science knowledge system graph based on the pre-constructed science knowledge system graph and the user behavior sample set of the target user, and to generate a cognitive state feature vector. It also uses a pre-constructed clustering algorithm to divide all active users of the science platform into at least one learning community and its group cognitive level vector, and calculates the knowledge influence weight of the target user in its learning community. The matching module is used to obtain a comprehensive matching degree for each candidate popular science knowledge content item based on its associated popular science knowledge dimensions and content characteristics, combined with the cognitive state feature vector and knowledge influence weight of the target user. The generation module is used to sort all candidate popular science knowledge content items in descending order according to the comprehensive matching degree, generate a target push list, and push it to the target user terminal.
[0055] It should be noted that other specific implementation details of the embodiments of the present invention can refer to the above-described method for accurately pushing popular science knowledge.
[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for accurately delivering popular science knowledge, characterized in that, include: S101, acquire the interactive behavior data of the target users on the science popularization knowledge platform within a preset time period, and form a time-series user behavior sample set; S102, based on the pre-constructed popular science knowledge system map and the user behavior sample set of the target users, use the preset cognitive assessment algorithm to evaluate the cognitive level score of the target users on at least one popular science knowledge dimension defined in the popular science knowledge system map, and generate a cognitive state feature vector. S103: Using a pre-defined clustering algorithm, all active users of the popular science knowledge platform are divided into at least one learning community. Each learning community corresponds to a group cognitive level vector. The knowledge influence weight of the target user in the learning community is calculated. S104. For each candidate popular science knowledge content item, based on its associated popular science knowledge dimensions and content characteristics, combined with the cognitive state feature vector and knowledge influence weight of the target user, a comprehensive matching degree is obtained. S105: Sort all candidate science popularization content items in descending order according to the overall matching degree, generate a target push list, and push it to the target user terminal.
2. The method for accurately pushing popular science knowledge as described in claim 1, characterized in that, The method for constructing the aforementioned popular science knowledge system map is as follows: We acquire popular science content from popular science knowledge platforms, extract the popular science knowledge dimension and difficulty level value tags for each popular science knowledge content item, and construct a triplet knowledge graph with the structure of "popular science knowledge dimension - connection edge - content item" based on the connection relationship between popular science knowledge dimensions.
3. The method for accurately pushing popular science knowledge as described in claim 2, characterized in that, S102 specifically includes: Based on the user behavior sample set, obtain the interaction behavior data of the target users on each predefined dimension of popular science knowledge in the popular science knowledge system map; Based on content interaction metrics from interactive behavior data, a pre-defined cognitive state assessment model is used to calculate the cognitive level score of the target user in at least one predefined science knowledge dimension. This process iterates through the science knowledge dimensions in the science knowledge system graph to generate the user's cognitive state feature vector. , Let m be the cognitive level score for the i-th knowledge dimension, and m be the total number of predefined knowledge dimensions in the graph. .
4. The method for accurately pushing popular science knowledge as described in claim 3, characterized in that, The preset cognitive assessment algorithm is set to perform a weighted fusion of average reading time, average answer accuracy, and content review rate under the corresponding knowledge dimension: The average reading time, average quiz accuracy, and content revisit rate were all obtained from a user behavior sample set. This represents the average reading time of the target user within knowledge dimension i. The average correct answer rate of the target users in knowledge dimension i. This indicates the user's content replay rate under knowledge dimension i. For the weight parameters, satisfying , This is the normalized benchmark value for the corresponding indicator, used for normalization processing.
5. The method for precisely delivering popular science knowledge as described in claim 4, characterized in that, The process of using a pre-defined clustering algorithm to divide all active users of the science popularization platform into at least one learning community and its group cognitive level vector specifically includes: Based on the cognitive state feature vectors of all active users, a pre-defined clustering algorithm is used to cluster them to generate at least one learning community. Each learning community includes at least one active user and its cognitive state feature vector.
6. The method for accurately pushing popular science knowledge as described in claim 4, characterized in that, The group cognitive level vector of the learning community is set as follows: the average cognitive level score of each active user on each knowledge dimension is calculated, which is the average feature vector of the cognitive state feature vector of all active users, forming a group cognitive level vector that represents the overall knowledge level of the community. .
7. The method for precisely delivering popular science knowledge as described in claim 5, characterized in that, The comprehensive matching degree is obtained by combining the cognitive state feature vector of the target user and the knowledge influence weight within their learning community, specifically including: Calculate the first matching score between each candidate science popularization knowledge content item and the cognitive state feature vector of the target user, and the second matching score between each candidate science popularization knowledge content item and the knowledge influence weight of the target user's community. Then, weight and fuse the first matching score and the second matching score to obtain the comprehensive matching degree.
8. The method for accurately pushing popular science knowledge as described in claim 7, characterized in that, The second matching score is calculated based on the second matching algorithm. : in, The knowledge influence factor of target user a within their community g. Let k be the set of science knowledge dimensions associated with the candidate science knowledge content item k. This represents the number of science knowledge dimensions associated with candidate science knowledge content item k. The average cognitive level score of community g on the knowledge dimension c. This represents the learning popularity of content within the knowledge dimension c in community g within a preset time window. and These are the preset balance coefficients. This is used to adjust the relative importance of community level fit and community learning enthusiasm.
9. The method for accurately pushing popular science knowledge as described in claim 8, characterized in that, Specifically, S105 includes: The overall matching degree of all candidate popular science knowledge content items is sorted in descending order to generate an initial recommendation sequence. The top three content items in the initial recommendation sequence are selected to form the target push list.
10. A system for precisely pushing popular science knowledge, applied to a method for precisely pushing popular science knowledge as described in any one of claims 1 to 9, characterized in that, The system includes: an acquisition module, an evaluation module, a matching module, and a generation module; The acquisition module is used to acquire interactive behavior data of target users within a preset time period on the popular science knowledge platform, forming a time-series user behavior sample set. The evaluation module is used to evaluate the cognitive level score of the target user in at least one science knowledge dimension defined in the science knowledge system graph based on the pre-constructed science knowledge system graph and the user behavior sample set of the target user, and to generate a cognitive state feature vector. It also uses a pre-constructed clustering algorithm to divide all active users of the science platform into at least one learning community and its group cognitive level vector, and calculates the knowledge influence weight of the target user in its learning community. The matching module is used to obtain a comprehensive matching degree for each candidate popular science knowledge content item based on its associated popular science knowledge dimensions and content characteristics, combined with the cognitive state feature vector and knowledge influence weight of the target user. The generation module is used to sort all candidate popular science knowledge content items in descending order according to the comprehensive matching degree, generate a target push list, and push it to the target user terminal.