A scientific and technological achievement pushing method based on an improved multi-granularity rough set

Abstract

This invention discloses a method for recommending scientific and technological achievements based on an improved multi-granularity rough set model, relating to the fields of intelligent information recommendation and data mining. The method includes: acquiring a dataset of scientific and technological achievements and constructing a multi-dimensional feature vector space by introducing a dynamic granularity adjustment factor for granularity partitioning; constructing an improved multi-granularity rough set model with an adaptive boundary adjustment mechanism based on the multi-dimensional feature vector space; calculating the upper and lower approximate sets of user historical behavior in the model to generate a fuzzy boundary region for user needs; calculating the multi-granularity matching degree based on the set of feature data of scientific and technological achievements and the fuzzy boundary region of user needs, and generating a personalized list of recommended scientific and technological achievements accordingly, outputting the final recommendation scheme. This invention can effectively characterize the uncertainty of user needs and the relationship between multi-granularity features, improving the accuracy and adaptability of scientific and technological achievement recommendations.

Description

Technical Field

[0001] This invention relates to the field of intelligent information recommendation technology in the transformation of scientific and technological achievements, and in particular to a method for pushing scientific and technological achievements based on improved multi-granularity rough sets. Background Technology

[0002] With the rapid growth in the number of scientific and technological achievements, universities, research institutes, and enterprises are placing higher demands on accurate and efficient information delivery mechanisms in the processes of technology transfer and technology supply-demand matching. Existing technology delivery technologies typically rely on keyword matching, collaborative filtering, or semantic similarity calculation to analyze the correlation between achievement information and user needs, which to some extent alleviates the problems of low efficiency and strong subjectivity associated with manual screening. However, scientific and technological achievements themselves are characterized by multiple technical dimensions, complex structures, and highly uncertain application scenarios, while user needs exhibit characteristics such as phased changes, ambiguous preferences, and behavioral evolution. This makes it difficult for traditional delivery methods based on single granularity or static feature spaces to accurately depict the multi-level relationship between "demand and achievement." Especially when user needs are not clearly defined and behavioral samples are incomplete, existing methods generally lack the ability to systematically model uncertainty and ambiguity, easily leading to problems such as highly homogenized recommendation results, insufficient relevance, or poor adaptability, thus restricting the application effect of technology delivery technologies in actual transformation scenarios.

[0003] CN121213197A discloses an intelligent recommendation method and system for technology transfer. This scheme uses a data processing module to uniformly collect and preprocess achievement data, enterprise demand data, and third-party data. It then utilizes knowledge graph technology to construct semantic relationships between achievements and demands. In the intelligent recommendation stage, it comprehensively calculates semantic relevance, semantic matching, and collaborative filtering, and generates a comprehensive recommendation score through weighted fusion to achieve the ranking and push of technology achievements. This technology enhances the semantic understanding and global relevance of recommendation results to some extent. However, its core reliance on knowledge graphs and multi-index weighted models means that the expression of user needs remains primarily at the level of explicit features and static semantics, failing to model the ambiguity and hierarchical differences of needs implicit in users' historical behavior. Furthermore, this method lacks a mechanism to characterize the dynamic changes in decision boundaries at different feature granularities. When user needs evolve over time or exhibit cross-domain, multi-scale characteristics, the adaptability and boundary discrimination accuracy of the recommendation model remain limited.

[0004] CN104036022A proposes a personalized recommendation method based on extended rough sets with variable precision tolerance relations. By constructing variable precision tolerance relations, it analyzes the indistinguishability of objects and extracts an effective set of rules by combining a discriminative matrix and attribute reduction mechanisms to support personalized recommendation decisions. This method introduces the concept of variable precision within the framework of rough set theory, which helps reduce the impact of noisy data on the rule extraction process and improves the stability of recommendation rules to some extent. However, this technique mainly focuses on tolerance relations at a single precision level, and its granularity structure is relatively fixed, making it difficult to adapt to the multi-scale distribution characteristics of scientific and technological achievement data in a multi-dimensional feature space. Furthermore, this method relies heavily on precise matching of conditional attributes to characterize user needs, lacks explicit modeling of ambiguous boundary regions of needs, and does not provide a cross-granularity level boundary adaptive adjustment mechanism. Therefore, the recommendation effect still has room for further improvement when facing scenarios with unclear user needs or sparse behavioral features.

[0005] To address the aforementioned shortcomings, this invention provides a method for pushing scientific and technological achievements based on an improved multi-granularity rough set. This method utilizes the correlation characteristics between user historical behavior data and scientific and technological achievement feature data at multiple granular levels to achieve refined calculation of the matching degree between scientific and technological achievements and user needs. Summary of the Invention

[0006] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0007] In view of the problems that existing technology delivery technologies generally suffer from in the process of user demand modeling, such as single granularity, insufficient characterization of demand boundaries, and limited adaptability to the uncertainty of user behavior, this invention is proposed.

[0008] Therefore, the problem to be solved by this invention is how to accurately characterize the fuzzy boundaries of user needs in a multi-dimensional feature space through multi-granularity modeling and adaptive boundary adjustment mechanisms, and achieve highly matched personalized push between scientific and technological achievements and user needs.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0010] In a first aspect, embodiments of the present invention provide a method for pushing scientific and technological achievements based on an improved multi-granularity rough set, comprising,

[0011] A dataset of scientific and technological achievements is obtained. The dataset is divided into granularities by introducing a dynamic granularity adjustment factor, and a multidimensional feature vector space is constructed. The dataset of scientific and technological achievements includes a set of user historical behavior data and a set of scientific and technological achievement feature data.

[0012] Based on the multidimensional feature vector space, an improved multi-granularity rough set model is constructed. The improved multi-granularity rough set model dynamically adjusts the boundary region range of different granularity levels by introducing an adaptive boundary adjustment mechanism.

[0013] Calculate the upper and lower approximate sets of each element in the user's historical behavior data set in the improved multi-granularity rough set model to generate the fuzzy boundary region of user needs;

[0014] Based on the data set of scientific and technological achievements and the fuzzy boundary region of user needs, the multi-granularity matching degree between scientific and technological achievements and user needs is calculated, and a personalized list of scientific and technological achievements is generated through the granularity division results, and the final recommendation scheme is output.

[0015] Secondly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any step of the above-described method for pushing scientific and technological achievements based on improved multi-granularity rough sets.

[0016] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing a computer program thereon, wherein: when the computer program is executed by a processor, it implements any step of the above-described method for pushing scientific and technological achievements based on improved multi-granularity rough sets.

[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: By introducing a dynamic granularity adjustment factor to divide the dataset of scientific and technological achievements into granular parts and constructing a multi-dimensional feature vector space, the features of scientific and technological achievements and the features of users' historical behavior can be adaptively expressed at different scales, avoiding information redundancy or feature loss caused by fixed granularity; on this basis, an improved multi-granularity rough set model with an adaptive boundary adjustment mechanism is constructed to achieve dynamic characterization of decision boundaries at different granularity levels, enhancing the model's ability to distinguish complex data distributions and uncertain samples; by calculating the upper approximation set and lower approximation set of users' historical behavior in the multi-granularity rough set model, a fuzzy boundary region of user needs is generated, realizing explicit modeling of users' implicit needs and needs uncertainty; based on the fuzzy boundary of scientific and technological achievements features and user needs, the corrected multi-granularity matching degree is calculated and a personalized push list is generated, so that the recommendation results reflect both the degree of needs matching and the stability of needs, thereby improving the accuracy, adaptability and practical application value of scientific and technological achievements push as a whole. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0019] Figure 1 The flowchart shows a method for pushing scientific and technological achievements based on an improved multi-granularity rough set model.

[0020] Figure 2 This is a user terminal interface diagram illustrating the improved multi-granularity rough set-based method for pushing scientific and technological achievements. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0022] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] As mentioned in the background section, existing technology recommendation methods generally suffer from problems such as a single granularity level, unclear demand boundaries, and insufficient adaptability to user behavior uncertainties during user demand modeling. These issues make it difficult to accurately reflect the complex relationship between the characteristics of technological achievements and users' actual needs. To address these problems, this invention provides a technology recommendation method based on an improved multi-granularity rough set approach.

[0025] Reference Figure 1 , Figure 1 This is a flowchart illustrating a method for pushing scientific and technological achievements based on an improved multi-granularity rough set according to an embodiment of the present invention. Figure 1 As shown, a method for pushing scientific and technological achievements based on improved multi-granularity rough sets includes:

[0026] S1: Obtain the dataset of scientific and technological achievements, divide the dataset into granularities by introducing a dynamic granularity adjustment factor, and construct a multi-dimensional feature vector space. The dataset of scientific and technological achievements includes a set of user historical behavior data and a set of scientific and technological achievement feature data.

[0027] Specifically, raw data is collected from the database of the scientific and technological achievements management platform to construct a scientific and technological achievements dataset; the user historical behavior data set includes browsing history, download history, and tags of areas of interest; the scientific and technological achievements feature data set includes technical keywords, application fields, and achievement types; and the dynamic granularity adjustment factor is adaptively adjusted according to the frequency of user historical interactions and the similarity of achievement features.

[0028] In an optional embodiment, suppose that in a certain technology achievement management platform, user A recently browsed 5 achievements related to "artificial intelligence chip design", downloaded 2 of them, and followed the two field tags "integrated circuit" and "neural network acceleration".

[0029] S1.1: Perform structured processing on the user's historical behavior data set and extract multiple behavioral feature fields from the user's historical behavior data set;

[0030] S1.2: Perform attribute parsing on the set of characteristic data of scientific and technological achievements, and extract multiple characteristic attribute fields from the set of characteristic data of scientific and technological achievements;

[0031] It should be noted that the various behavioral feature fields include browsing history, download history, and interest area tags. Browsing history records the timestamps and duration of user visits to the scientific and technological achievement pages; download history records the user's operation log for obtaining the full text of the scientific and technological achievement; interest area tags record the technical interest directions actively marked by the user; and various feature attribute fields include technical keywords, application fields, and achievement types. Technical keywords are the set of core technical terms of the scientific and technological achievement, application fields are the industry application classification identifiers of the scientific and technological achievement, and achievement types are the form category tags of the scientific and technological achievement.

[0032] S1.3: Based on multiple behavioral feature fields, count the number of times each user interacts with various scientific and technological achievements within a preset time window, and calculate the user's historical interaction frequency. The specific formula is as follows:

[0033] ;

[0034] in, This represents the frequency of user history interactions within a preset time window. The total number of technological achievements related to user interaction. For indexing scientific and technological achievements, For browsing behavior weighting coefficients, Let j be the number of times a user views the j-th result. For download behavior weighting coefficients, Let j be the number of times a user downloads the j-th result. To focus on the tag weight coefficient, This is the matching identifier between the j-th result and the user's area of ​​interest (1 for a match, 0 otherwise). The time decay coefficient, For the current time, The time taken for the user to interact with the j-th result.

[0035] It should be noted that the frequency of user historical interactions Range description: A higher value indicates a higher level of user interaction activity, reflecting the frequency of historical user interactions. A value less than 5 indicates a low-activity user; a value between 5 and 5 indicates the frequency of the user's historical interactions. Users with a frequency of less than 20 hours are considered moderately active, based on their historical interaction frequency. Users with ≥20 active users are considered highly active.

[0036] S1.4: Based on multiple feature attribute fields, a similarity algorithm is used to calculate the feature similarity between any two scientific and technological achievements. The specific formula is as follows:

[0037] ;

[0038] in, The similarity of achievement characteristics between scientific and technological achievement i and scientific and technological achievement j. For the similarity weight of technical keywords, For keyword vector dimensions, For keyword indexing, Let i be the weight value of scientific and technological achievement i in the k-th keyword. Let j be the weight value of the scientific and technological achievement in the k-th keyword. For application domain similarity weights, It is a collection of application fields of scientific and technological achievement i. This refers to the collection of application areas of scientific and technological achievement j. As the weight of similarity between result types, Encoding the type of scientific and technological achievement i, Encode the type of scientific and technological achievement j. For result type matching function (1 for the same type, 0 otherwise), where .

[0039] It should be noted that the similarity of the results characteristics Range description: ,when A similarity of <0.3 indicates low similarity, while 0.3 ≤ A similarity of <0.7 indicates moderate similarity. A similarity score of ≥0.7 indicates high similarity.

[0040] S1.5: Based on the frequency of users' historical interactions and the similarity of their results, a dynamic granularity adjustment factor is constructed through a nonlinear mapping function. The specific formula is as follows:

[0041]

[0042] ;

[0043] in, For user u, the dynamic granularity adjustment factor The interaction frequency response coefficient, This is a normalized value for the frequency of user interactions. This is the interaction frequency threshold parameter. The similarity response coefficient, This is the normalized value of the average similarity of user interaction outcomes. For similarity threshold parameters, The average feature similarity of the set of historical interaction results of user u. and These are the minimum and maximum interaction frequencies, respectively. and These are the minimum and maximum values ​​of feature similarity, respectively.

[0044] It should be noted that the dynamic granularity adjustment factor is adaptively adjusted based on the normalized value of the user's historical interaction frequency and the mean normalized value of the similarity of the outcome features. Range: ,when When the value is ≤0.3, a coarse-grained partitioning strategy is adopted; when 0.3 < When <0.7, a medium-granularity partitioning strategy is adopted. When the value is ≥0.7, a fine-grained partitioning strategy is adopted.

[0045] S1.6: Obtain the dynamic granularity adjustment factor, compare the dynamic granularity adjustment factor with the preset granularity threshold, and obtain the set of scientific and technological achievement features. The set of scientific and technological achievement features includes a fine-grained set of scientific and technological achievement features and a coarse-grained set of scientific and technological achievement features, specifically including:

[0046] When the value of the dynamic granularity adjustment factor exceeds the preset granularity threshold, a fine-grained partitioning strategy is adopted to perform feature decomposition on the scientific and technological achievement dataset, generating a fine-grained set of scientific and technological achievement features. For each scientific and technological achievement in the dataset, technical keywords are extracted from multiple feature attribute fields. These technical keywords are then split into primary keywords and secondary keywords according to semantic hierarchy. Primary keywords represent the top-level concepts of the technical field, while secondary keywords represent specific technical terms. For application fields, the application fields are divided into main application fields and sub-application fields according to industry classification standards. The main application field represents the main application industry of the scientific and technological achievement, while the sub-application field represents the subdivided application scenarios of the scientific and technological achievement. For achievement types, the achievement types are divided into basic achievement types and extended achievement types according to the subdivision standards of achievement forms. Basic achievement types represent the basic form classification of scientific and technological achievements, while extended achievement types represent the additional attribute classification of scientific and technological achievements. The primary keywords, secondary keywords, main application fields, sub-application fields, basic achievement types, and extended achievement types are used as fine-grained feature attribute fields to complete the feature decomposition operation of the scientific and technological achievement dataset, generating a fine-grained set of scientific and technological achievement features.

[0047] When the value of the dynamic granularity adjustment factor is less than or equal to the preset granularity threshold, a coarse-grained partitioning strategy is adopted to perform feature aggregation on the scientific and technological achievement dataset, generating a coarse-grained set of scientific and technological achievement features. Multiple scientific and technological achievements in the dataset are clustered according to the semantic similarity of their technical keywords. Technical keywords with semantic similarity greater than the preset similarity threshold are grouped into unified keyword clusters, each identified by a representative keyword. Application fields are also merged and integrated according to their industry categories, combining multiple application fields belonging to the same major industry category into unified application field categories, each identified by an industry category label. Furthermore, achievement types are categorized and merged according to their formal attributes, integrating multiple achievement types with similar formal attributes into unified achievement type groups, each identified by a type group label. The representative keywords, industry category labels, and type group labels are used as coarse-grained feature attribute fields to complete the feature aggregation operation on the scientific and technological achievement dataset, generating a coarse-grained set of scientific and technological achievement features.

[0048] For the fine-grained set of scientific and technological achievements features, a fine-grained weight value is assigned to each fine-grained feature attribute field. The fine-grained weight value is assigned based on the frequency of occurrence of the fine-grained feature attribute field in the user's historical behavior data set. The fine-grained feature attribute field with a higher frequency of occurrence corresponds to a larger fine-grained weight value. For the coarse-grained set of scientific and technological achievements features, a coarse-grained weight value is assigned to each coarse-grained feature attribute field. The coarse-grained weight value is assigned based on the number of scientific and technological achievements covered by the coarse-grained feature attribute field. The coarse-grained feature attribute field that covers more scientific and technological achievements corresponds to a larger coarse-grained weight value.

[0049] The fine-grained and coarse-grained sets of scientific and technological achievements are labeled and stored. The fine-grained set of scientific and technological achievements is labeled as a high-discrimination feature set, and the coarse-grained set of scientific and technological achievements is labeled as a high-coverage feature set, thus completing the granularity division operation based on the dynamic granularity adjustment factor.

[0050] S1.7: Vectorize the set of features of scientific and technological achievements to construct a multi-dimensional feature vector space;

[0051] S1.7.1: Obtain the feature set of scientific and technological achievements, perform a full-domain scan and deduplication merging of all feature attribute fields contained in the feature set of scientific and technological achievements, and generate a unified full-domain feature dictionary;

[0052] S1.7.2: For each scientific and technological achievement in the set of scientific and technological achievements features, vector representation is performed using one-hot encoding or term frequency-inverse document frequency method based on the global feature dictionary;

[0053] Furthermore, the various feature attribute fields of the current scientific and technological achievements are mapped to the corresponding weight values ​​in the global feature dictionary, thereby transforming the scientific and technological achievements into a high-dimensional sparse feature vector.

[0054] S1.7.3: Perform behavior-based feature vectorization on user historical behavior data;

[0055] Furthermore, based on the user's historical behavior data set, the frequency of interaction between the target user and scientific and technological achievements with different characteristic attribute fields is statistically analyzed; the interaction frequency is used as a weight to perform weighted accumulation and normalization on the feature vector of the scientific and technological achievements, generating a comprehensive user feature vector representing the user's long-term interests and preferences.

[0056] S1.7.4: Project the feature vectors of all scientific and technological achievements and the comprehensive feature vectors of users together into a space with the various dimensions of the global feature dictionary as coordinate axes;

[0057] Preferably, in this space, each scientific and technological achievement and each user is represented by a definite feature vector point, and this space is called the multidimensional feature vector space.

[0058] For example, historical behavioral data of user A (such as browsing time and duration, download logs, and followed tags) and feature data of all scientific and technological achievements (including technical keywords, application fields, and achievement types) are collected. Based on user A's interaction records in the past month, their historical interaction frequency is calculated: assuming a browsing behavior weight of 0.3, a download behavior weight of 0.5, and a followed tag matching weight of 0.2, and incorporating time decay (e.g., higher weight for recent behavior), the calculated historical interaction frequency is 18, classifying them as a moderately active user. Simultaneously, the platform calculates the feature similarity between any two scientific and technological achievements. For example, two achievements related to "GPU acceleration" and "FPGA optimization," respectively, are compared based on their technical keyword weights, application field overlap, and achievement type consistency. A weighted similarity of 0.65 is obtained, classifying them as moderately similar. Based on the historical interaction frequency and the mean similarity, the platform generates a dynamic granularity adjustment factor for the user using a non-linear mapping function, resulting in a value of 0.6, corresponding to a medium-granularity segmentation strategy. Therefore, the system performs medium-granularity feature processing on the scientific and technological achievement dataset: for example, it splits technical keywords into primary keywords "artificial intelligence" and secondary keywords "chip design" according to semantic hierarchy; it subdivides the application field into the main field "electronic information" and the subfield "integrated circuits"; and it distinguishes between the basic type "patent" and the extended type "prototype system" according to the form of the achievement. These processed features are vectorized to construct a multi-dimensional feature vector space.

[0059] S2: Based on the multidimensional feature vector space, an improved multi-granularity rough set model is constructed. The improved multi-granularity rough set model dynamically adjusts the boundary region range of different granularity levels by introducing an adaptive boundary adjustment mechanism.

[0060] S2.1: Based on the multidimensional feature vector space, a decision information system is constructed by taking all feature vector points in the multidimensional feature vector space as the universe of discourse and the attributes of each dimension as the set of condition attributes. The decision information system includes the universe of discourse, the set of condition attributes, and decision attributes.

[0061] S2.2: Set a roughness threshold based on the number of attributes and the distribution of attribute values ​​in the condition attribute set, and set a fine-grained weight coefficient based on the distinguishing ability of each attribute in the classification decision.

[0062] Specifically, the roughness threshold is used to define the acceptable range of the boundary region. When the uncertainty of the membership of an object exceeds the roughness threshold, the object is classified into the boundary region for subsequent processing.

[0063] It should be noted that the fine-grained weight coefficient assigns differentiated weight values ​​to each conditional attribute; attributes with stronger distinguishing ability receive higher fine-grained weight coefficients, while attributes with weaker distinguishing ability receive lower fine-grained weight coefficients.

[0064] S2.3: Based on the set of conditional attributes and fine-grained weight coefficients, construct weighted indistinguishable relations of multiple granular levels on the universe of discourse, including coarse-grained, medium-grained and fine-grained levels.

[0065] Furthermore, the weighted indistinguishable relation introduces fine-grained weight coefficients into the calculation process of the traditional indistinguishable relation, making the impact of high-weight attributes on the object's distinguishability greater than that of low-weight attributes. Multiple granularity levels include coarse-grained, medium-grained, and fine-grained levels.

[0066] S2.4: Divide the universe of discourse according to the weighted indistinguishability relation to generate a set of equivalence classes at this granularity level, wherein each equivalence class in the set of equivalence classes contains objects with the same attribute characteristics at this granularity level.

[0067] S2.5: An adaptive boundary adjustment mechanism is introduced, which dynamically calculates the boundary adjustment factor based on the roughness threshold and the distribution density of the equivalence class set. The specific formula is as follows:

[0068]

[0069] ;

[0070] in, The boundary adjustment factor for the l-th granularity level. Let l be the equivalence class distribution density of the l-th granularity level. The reference density constant, To preset the roughness threshold, This represents the actual roughness value at the l-th grain size level. For roughness sensitivity parameters, Let be the number of equivalence classes at the l-th granularity level. Let be the volume of the feature space at the l-th granularity level.

[0071] It should be noted that the boundary adjustment factor Range: ,when When the value is greater than 0.5, the boundary region narrows. When <0.5, the boundary region expands. When the value is 0.5, the boundary region remains unchanged.

[0072] Furthermore, the boundary adjustment factor is used to differentiate the boundary region range at different granularity levels. If the distribution density of the equivalence class set is high, the boundary adjustment factor narrows the boundary region range; if the distribution density of the equivalence class set is low, the boundary adjustment factor expands the boundary region range.

[0073] S2.6: Based on the boundary adjustment factor, calculate the adjusted boundary region range for multiple granularity levels, and generate the dynamic boundary region corresponding to each granularity level. The range of the dynamic boundary region is adaptively adjusted according to the distribution characteristics of the equivalence class set.

[0074] S2.6.1: For each level in multiple granularity levels, obtain the set of equivalence classes divided by the weighted indistinguishable relation, and calculate the distribution density of the equivalence class set in the universe of discourse at the current level;

[0075] S2.6.2: Based on the boundary adjustment factor, combined with the distribution density of the current granularity level and the preset roughness threshold, calculate the boundary expansion or boundary contraction amount corresponding to the level.

[0076] S2.6.3: Adjust the range of the original boundary region based on the boundary expansion or contraction amount. Specifically, this includes: proportionally including or excluding objects in the original boundary region according to their characteristic distances to objects in the lower and upper approximation sets of their nearest neighbors, in order to form an adjusted set of boundary objects.

[0077] S2.6.4: The adjusted boundary object set of each granularity level is recombined with the inherent lower approximation set and upper approximation set of that level to jointly constitute the final dynamic boundary region of that level. The range and specific object composition of the dynamic boundary region are jointly determined by the boundary adjustment factor, distribution density and roughness threshold.

[0078] S2.7: Integrate and encapsulate the decision information system, roughness threshold, fine-grained weight coefficient, weighted indistinguishable relation, adaptive boundary adjustment mechanism and dynamic boundary region to construct an improved multi-granularity rough set model. The improved multi-granularity rough set model supports approximate set calculation and boundary region analysis at the coarse-grained level, medium-grained level and fine-grained level.

[0079] In an optional embodiment, the platform establishes an improved multi-granularity rough set model based on the aforementioned vector space; it maps user behavior records and scientific and technological achievements to objects in the domain of discourse, uses features such as technical keywords and application fields as conditional attributes, and assigns weights based on the contribution of each attribute in distinguishing user interests (e.g., technical keywords have higher weights); through weighted indistinguishable relations, it divides objects at three granularity levels—coarse, medium, and fine—to form equivalence classes; for example, at the fine-granularity level, "convolutional neural network optimization" and "attention mechanism hardware implementation" are divided into different equivalence classes due to differences in technical details; while at the coarse-granularity level, they are grouped into the same "AI acceleration" category. The improved multi-granularity rough set model introduces an adaptive boundary adjustment mechanism: based on the distribution density of equivalence classes at each granularity level and a preset roughness threshold, the boundary adjustment factor is dynamically calculated. If the equivalence classes at a certain level are densely distributed (such as the clustering of similar technological achievements at the fine-grained level), the boundary adjustment factor is greater than 0.5, and the system will narrow the boundary region to improve recommendation accuracy. If the distribution is sparse (such as the mixing of cross-domain achievements at the coarse-grained level), the boundary adjustment factor is less than 0.5, and the system will expand the boundary region to cover a wider range of possibilities. The improved multi-granularity rough set model generates dynamic boundary regions at each granularity level, supporting multi-granularity approximation calculations and uncertainty analysis.

[0080] S3: Calculate the upper approximation set and lower approximation set of each element in the user historical behavior data set in the improved multi-granularity rough set model, and generate the user demand fuzzy boundary region, where the user demand fuzzy boundary region is used to quantify the degree of uncertainty of user demand;

[0081] S3.1: Extract elements one by one from the user historical behavior data set, map them to the universe of discourse of the improved multi-granularity rough set model, and generate a set of user behavior objects to be analyzed, where each object in the set of user behavior objects to be analyzed corresponds to a user historical behavior record.

[0082] S3.2: For each object in the set of user behavior objects to be analyzed, calculate the approximate set and the upper approximate set at multiple granularity levels, and calculate the difference between the upper approximate set and the lower approximate set to generate the original boundary region corresponding to each granularity level;

[0083] Specifically, based on the weighted indistinguishable relation and the equivalence class set, the lower approximation set and the upper approximation set are calculated. Objects that are completely contained in the equivalence class set of the target concept are included in the lower approximation set; objects that intersect with the target concept in the equivalence class set are included in the upper approximation set; the original boundary region contains objects whose belonging to the target concept cannot be clearly determined, and the category of objects in the original boundary region is uncertain.

[0084] S3.2.1: Based on the weighted indistinguishable relation constructed in the improved multi-granularity rough set model, for any target object in the set of user behavior objects to be analyzed, determine the set of equivalence classes to which the target object belongs at the coarse-grained level, the medium-grained level, and the fine-grained level respectively.

[0085] S3.2.2: Using the set of equivalence classes to which the target object belongs in each granularity level as the basic calculation unit, and combining the decision attributes defined in the improved multi-granularity rough set model, calculate the lower approximate set of the target object at each granularity level. The lower approximate set consists of all objects at that granularity level that are indistinguishable from the target object and have completely consistent decision attributes.

[0086] S3.2.3: Based on the set of equivalence classes to which the target object belongs, calculate the upper approximate set of the target object at each granularity level, wherein the upper approximate set consists of all objects at that granularity level that are indistinguishable from the target object and whose decision attributes are related.

[0087] S3.2.4: Based on the dynamic boundary regions generated for each granularity level by the improved multi-granularity rough set model, the upper approximation set and lower approximation set of the target object at each level are corrected. Specifically, objects belonging to the dynamic boundary region are removed from the lower approximation set and are kept in the upper approximation set.

[0088] S3.2.5: Aggregate the difference between the corrected upper approximation set and the lower approximation set of the target object at each granularity level to generate a multi-granularity fuzzy boundary set that represents the uncertainty of the target object's demand.

[0089] S3.2.6: Traverse all objects in the set of user behavior objects to be analyzed, repeat steps S3.2.1 to S3.2.5, summarize the multi-granularity fuzzy boundary set of all objects, and finally generate a comprehensive user demand fuzzy boundary region.

[0090] S3.3: Based on the adaptive boundary adjustment mechanism and boundary adjustment factor, the original boundary regions of each granularity level are corrected to generate the adjusted boundary regions corresponding to each granularity level.

[0091] S3.3.1: For any current granularity level among coarse-grained, medium-grained, and fine-grained levels, obtain the original boundary region and boundary adjustment factor of that level;

[0092] S3.3.2: Determine the type of correction operation for the original boundary region based on the value of the boundary adjustment factor and the preset adjustment direction rules;

[0093] Furthermore, when the value of the boundary adjustment factor is greater than 1, a boundary expansion operation is performed; when the value of the boundary adjustment factor is less than 1, a boundary contraction operation is performed; and when the value of the boundary adjustment factor is equal to 1, the original boundary region remains unchanged.

[0094] S3.3.3: Based on the correction operation type, perform range correction on the set of objects in the original boundary region;

[0095] Furthermore, for the boundary expansion operation, some non-boundary objects belonging to the current granularity level and whose distance in feature space from objects in the original boundary region is less than a preset distance threshold are included in the correction range to form an expanded candidate object set; for the boundary contraction operation, some objects in the original boundary region whose distance in feature space from objects in the clearly defined upper or lower approximate set at the current granularity level is greater than another preset distance threshold are excluded from the correction range to form a contracted candidate object set.

[0096] S3.3.4: Perform a secondary membership determination on the candidate object set with the inherent lower approximation set and upper approximation set of the current granularity level;

[0097] Specifically, only objects in the candidate object set that neither uniquely belong to the lower approximation set nor completely belong to the upper approximation set are formally designated as objects in the adjusted boundary region of the current granularity level.

[0098] S3.3.5: Record all boundary objects finally determined by step S3.3.4 for the current granularity level, and generate the adjusted boundary region corresponding to the granularity level;

[0099] S3.3.6: Traverse the coarse-grained, medium-grained, and fine-grained levels, and repeat steps S3.3.1 to S3.3.5 for each level to generate the adjusted boundary regions corresponding to each of the granularity levels.

[0100] S3.4: Based on the adjusted boundary region, a fuzzy membership quantification method is introduced to calculate the degree of membership of each object to the target concept. The specific formula is as follows:

[0101] ;

[0102] in, Let x be the fuzzy membership value of object x with respect to target concept X. The shortest distance from object x to the lower approximate set. Let x be the shortest distance from object x to the complement of the upper approximate set. To prevent extremely small positive numbers from being divided by zero, The mean of the boundary adjustment factor.

[0103] It should be noted that the fuzzy membership value ranges from zero to one, and the closer the fuzzy membership value is to one, the higher the probability that the object belongs to the target concept. The closer the fuzzy membership value is to zero, the lower the probability that the object belongs to the target concept.

[0104] S3.5: Based on fuzzy membership values, a multi-granularity fusion strategy is adopted to comprehensively integrate boundary information at multiple granularity levels to generate the fuzzy boundary region required by the user.

[0105] S3.5.1: Obtain the fuzzy membership value corresponding to each object in the coarse-grained, medium-grained, and fine-grained levels. For any target object in the set of user behavior objects to be analyzed, extract the coarse-grained fuzzy membership value of the target object at the coarse-grained level, the medium-grained fuzzy membership value at the medium-grained level, and the fine-grained fuzzy membership value at the fine-grained level. Combine the coarse-grained fuzzy membership value, the medium-grained fuzzy membership value, and the fine-grained fuzzy membership value to form a multi-level fuzzy membership vector for the target object.

[0106] S3.5.2: Based on the fine-grained weight coefficient, set granularity level fusion weights for coarse-grained, medium-grained, and fine-grained levels respectively. The granularity level fusion weights are assigned differentiated values ​​according to the ability of each granularity level to distinguish user needs.

[0107] Furthermore, fine-grained levels have larger granularity level fusion weights due to their detailed feature division and strong discriminative ability; coarse-grained levels have smaller granularity level fusion weights due to their coarse feature division and wide coverage; medium-grained levels have granularity level fusion weights between those of fine-grained and coarse-grained levels; and the sum of granularity level fusion weights is normalized to one.

[0108] S3.5.3: For the multi-level fuzzy membership vector of the target object, multiply the coarse-grained fuzzy membership value with the granularity level fusion weight corresponding to the coarse-grained level to obtain the coarse-grained weighted membership component; multiply the medium-grained fuzzy membership value with the granularity level fusion weight corresponding to the medium-grained level to obtain the medium-grained weighted membership component; multiply the fine-grained fuzzy membership value with the granularity level fusion weight corresponding to the fine-grained level to obtain the fine-grained weighted membership component.

[0109] S3.5.4: Summing the coarse-grained weighted membership components, the medium-grained weighted membership components, and the fine-grained weighted membership components yields the comprehensive fuzzy membership value of the target object. The comprehensive fuzzy membership value represents the overall degree of membership of the target object to the target concept from a multi-grained fusion perspective.

[0110] S3.5.5: Set the comprehensive membership degree determination threshold, and divide the comprehensive membership degree determination threshold into a lower determination threshold and an upper determination threshold;

[0111] S3.5.6: For each object in the set of user behavior objects to be analyzed, the comprehensive fuzzy membership value is compared with the lower and upper decision thresholds, and the object classification operation is performed.

[0112] Furthermore, if the comprehensive fuzzy membership value is less than the lower decision threshold, the object is classified into the set of explicit negative domain objects; if the comprehensive fuzzy membership value is greater than the upper decision threshold, the object is classified into the set of explicit positive domain objects; if the comprehensive fuzzy membership value is greater than or equal to the lower decision threshold and less than or equal to the upper decision threshold, the object is classified into the set of fuzzy boundary objects.

[0113] It should be noted that the lower judgment threshold is used to define objects that clearly do not belong to the target concept, while the upper judgment threshold is used to define objects that clearly belong to the target concept; objects in the set of clearly negative domain objects represent technological achievements that clearly do not meet user needs; objects in the set of clearly positive domain objects represent technological achievements that clearly meet user needs; and objects in the set of fuzzy boundary objects represent technological achievements whose demand attribution is uncertain.

[0114] S3.5.7: Extract all objects from the fuzzy boundary object set, and calculate the boundary uncertainty index for each object by combining the distribution location information of each object in the adjusted boundary region at multiple granularity levels;

[0115] Specifically, the boundary uncertainty index is quantified based on the frequency of an object's appearance in the adjusted boundary regions at each granularity level. If an object appears in the adjusted boundary regions at the coarse-grained, medium-grained, and fine-grained levels, the boundary uncertainty index of that object is relatively high; if an object appears only in the adjusted boundary regions at some granularity levels, the boundary uncertainty index of that object is relatively low.

[0116] S3.5.8: Sort the objects in the fuzzy boundary object set in descending order according to the value of the boundary uncertainty index, and associate the sorted object sequence with the comprehensive fuzzy membership value, boundary uncertainty index and multi-level fuzzy membership vector of each object to generate a structured user requirement fuzzy boundary region data structure.

[0117] S3.5.9: Encapsulate and mark the user requirement fuzzy boundary region data structure, which includes a set of fuzzy boundary objects, the comprehensive fuzzy membership value of each object, the boundary uncertainty index of each object, and the membership distribution information of each object at multiple granularity levels, thus completing the generation operation of the user requirement fuzzy boundary region.

[0118] S3.6: Calculate the uncertainty measure of user requirements based on the fuzzy boundary region of user requirements. The specific formula is as follows:

[0119] ;

[0120] ;

[0121] in, This is a measure of the uncertainty of user needs. The number of objects in the fuzzy boundary object set. The total number of objects in the universe of discourse. The coefficient of variation of the membership values ​​of fuzzy boundary objects. The mean of the membership values ​​of the fuzzy boundary objects. To prevent extremely small positive numbers from being divided by zero.

[0122] It should be noted that the user demand uncertainty metric... Range: ,when When <0.3, user requirements are clear; when 0.3≤ When <0.6, user demand exhibits moderate uncertainty. When the value is ≥0.6, the user's needs are vague and the push notification scope needs to be expanded.

[0123] Furthermore, the uncertainty metric for user needs is comprehensively evaluated based on the proportion of objects in the fuzzy boundary region of user needs and the dispersion of fuzzy membership values. The uncertainty metric for user needs is used to quantify the clarity and fuzziness of the expression of user needs.

[0124] In an optional embodiment, the platform maps user A's historical behavior objects (such as the 5 results they have viewed) to an improved multi-granularity rough set model. It calculates the lower approximation set (results that explicitly meet user needs) and upper approximation set (results that may meet user needs) for each object at three granularity levels, and uses dynamic boundary regions for correction to obtain adjusted boundary regions. For example, a result about "in-memory computing chips" is classified as lower approximation at the fine-granularity level due to a high degree of matching with technical keywords, but is in the boundary region at the coarse-granularity level due to its broad scope. The platform uses a fuzzy membership quantification method to calculate the degree of membership of each object to user A's target needs and integrates information from the three granularity levels to obtain a comprehensive fuzzy membership degree. Assuming user A's membership degree for some results is between 0.4 and 0.6, these results are categorized into a fuzzy boundary object set. Further, based on the proportion of objects in this set to the total number of objects, and the dispersion of membership values ​​within the set, a user demand uncertainty metric of 0.5 is calculated, indicating that user A's needs have moderate uncertainty and their interest boundaries are relatively fuzzy.

[0125] S4: Based on the data set of scientific and technological achievements and the fuzzy boundary region of user needs, calculate the multi-granularity matching degree between scientific and technological achievements and user needs, sort them according to the numerical value of the multi-granularity matching degree, generate a personalized scientific and technological achievement push list through the granularity division results, and output the final recommendation scheme.

[0126] S4.1: Extract scientific and technological achievement objects one by one from the set of scientific and technological achievement feature data, and vectorize the technical keywords, application fields and achievement types of each scientific and technological achievement object to generate a set of scientific and technological achievement feature vectors;

[0127] S4.2: Based on multiple granularity levels, the granularity matching degree between the feature vector of scientific and technological achievements and the fuzzy boundary region of user needs is calculated according to the corresponding equivalence class set and the adjusted boundary region, generating a multi-granularity matching degree. The specific formula is as follows:

[0128] ;

[0129] in, For the multi-granularity matching degree of scientific and technological achievements s, This represents the total number of granularity levels. For granular level indexing, The fusion weight for the l-th granularity level is... The boundary matching degree of scientific and technological achievement s at the l-th granularity level. The standard deviation of the matching degree at each level. Let be the mean of the boundary matching degree of scientific and technological achievement s at the l-th granularity level. This is a measure of the uncertainty of user needs. To prevent extremely small positive numbers from being divided by zero.

[0130] It should be noted that the corrected multi-granularity matching degree of scientific and technological achievement s Range: ,when When the result s is ≥0.7, results that highly match user needs should be prioritized for push; when 0.4 ≤ When <0.7, it is a moderate match. When the match value is 0.4, the matching degree is low and it is not recommended to push.

[0131] S4.2.1: Obtain any target scientific and technological achievement feature vector from the set of scientific and technological achievement feature vectors, and map the target scientific and technological achievement feature vector to the feature spaces corresponding to the coarse-grained level, medium-grained level and fine-grained level in the improved multi-grained rough set model respectively;

[0132] Preferably, based on the technical keywords, application fields, and achievement type attribute values ​​in the feature vector of the target scientific and technological achievement, the coarse-grained equivalence class to which the feature vector of the target scientific and technological achievement belongs is determined at the coarse-grained level, the medium-grained equivalence class to which the feature vector of the target scientific and technological achievement belongs is determined at the medium-grained level, and the fine-grained equivalence class to which the feature vector of the target scientific and technological achievement belongs is determined at the fine-grained level.

[0133] S4.2.2: For the coarse-grained level, extract all user behavior objects belonging to the adjusted boundary region of the coarse-grained level from the user demand fuzzy boundary region data structure, and calculate the equivalence class similarity between the coarse-grained equivalence class to which the feature vector of the target scientific and technological achievement belongs and the coarse-grained equivalence class to which each user behavior object belongs.

[0134] Furthermore, equivalence class similarity is calculated using cosine similarity or Euclidean distance algorithms. The smaller the distance between the center vectors of two equivalence classes in the feature space, the larger the equivalence class similarity value. Based on equivalence class similarity, the coarse-grained object similarity set of each user behavior object in the adjusted boundary region of the target scientific and technological achievement feature vector is calculated. All similarity values ​​in the coarse-grained object similarity set are weighted and averaged to obtain the coarse-grained boundary matching degree of the target scientific and technological achievement feature vector at the coarse-grained level.

[0135] It should be noted that the weighting coefficients in the weighted average operation are assigned based on the comprehensive fuzzy membership value of each user behavior object in the data structure of the fuzzy boundary region of user demand. The larger the comprehensive fuzzy membership value, the larger the weighting coefficient of the user behavior object.

[0136] S4.2.3: For the medium-granularity level, extract all user behavior objects belonging to the adjusted boundary region of the medium-granularity level from the user demand fuzzy boundary region data structure, and calculate the equivalence class similarity between the medium-granularity equivalence class to which the feature vector of the target scientific and technological achievement belongs and the medium-granularity equivalence class to which each user behavior object belongs.

[0137] Specifically, based on equivalence class similarity, the similarity set between the feature vector of the target scientific and technological achievement and the medium-granularity object of each user behavior object in the adjusted boundary region at the medium-granularity level is calculated. The weighted average of all similarity values ​​in the medium-granularity object similarity set is then used to obtain the medium-granularity boundary matching degree of the feature vector of the target scientific and technological achievement at the medium-granularity level.

[0138] S4.2.4: For the fine-grained level, extract all user behavior objects belonging to the adjusted boundary region of the fine-grained level from the data structure of the user demand fuzzy boundary region, and calculate the equivalence class similarity between the fine-grained equivalence class to which the feature vector of the target scientific and technological achievement belongs and the fine-grained equivalence class to which each user behavior object belongs.

[0139] Furthermore, based on equivalence class similarity, the fine-grained object similarity set of each user behavior object in the adjusted boundary region of the target scientific and technological achievement feature vector is calculated. The weighted average of all similarity values ​​in the fine-grained object similarity set is then used to obtain the fine-grained boundary matching degree of the target scientific and technological achievement feature vector at the fine-grained level.

[0140] S4.2.5: Extract coarse-grained boundary matching degree, medium-grained boundary matching degree and fine-grained boundary matching degree, and calculate the comprehensive granularity matching degree of the target scientific and technological achievement feature vector by combining the granularity level fusion weight;

[0141] Specifically, the coarse-grained boundary matching degree is multiplied by the granularity level fusion weight corresponding to the coarse-grained level to obtain the coarse-grained matching degree component; the medium-grained boundary matching degree is multiplied by the granularity level fusion weight corresponding to the medium-grained level to obtain the medium-grained matching degree component; the fine-grained boundary matching degree is multiplied by the granularity level fusion weight corresponding to the fine-grained level to obtain the fine-grained matching degree component; the coarse-grained matching degree component, the medium-grained matching degree component, and the fine-grained matching degree component are summed to obtain the comprehensive granularity matching degree.

[0142] S4.2.6: Introduce a granularity consistency verification mechanism to calculate the dispersion of the overall granularity matching degree and generate a granularity consistency coefficient;

[0143] It should be noted that the granularity consistency coefficient is quantitatively calculated based on the variance or standard deviation of the coarse-grained boundary matching degree, the medium-grained boundary matching degree, and the fine-grained boundary matching degree. When the boundary matching degree values ​​of the three granularity levels differ significantly, the granularity consistency coefficient is small, indicating that there is a significant discrepancy in the judgment of user needs at different granularity levels. When the boundary matching degree values ​​of the three granularity levels tend to be consistent, the granularity consistency coefficient is large, indicating that the judgment of user needs at different granularity levels tends to be unified.

[0144] S4.2.7: Based on the granularity consistency coefficient, adjust the confidence level of the comprehensive granularity matching degree to generate the multi-granularity matching degree of the feature vector of the target scientific and technological achievement;

[0145] Furthermore, the overall granularity matching degree is multiplied by the granularity consistency coefficient to obtain the multi-granularity matching degree. The granularity consistency coefficient is used as a confidence adjustment factor. When the granularity consistency coefficient is large, the multi-granularity matching degree is close to the original value of the overall granularity matching degree. When the granularity consistency coefficient is small, the multi-granularity matching degree is reduced relative to the original value of the overall granularity matching degree.

[0146] S4.3: Based on the uncertainty metric of user needs, perform confidence correction on the multi-granularity matching degree to generate the corrected multi-granularity matching degree;

[0147] Furthermore, the confidence of the multi-granularity matching degree is corrected based on the uncertainty metric of user demand. If the uncertainty metric of user demand is high, the confidence weight of the multi-granularity matching degree is reduced; if the uncertainty metric of user demand is low, the confidence weight of the multi-granularity matching degree is increased, thus generating the corrected multi-granularity matching degree.

[0148] S4.4: Sort the corrected multi-granularity matching degree in descending order of numerical value to generate a scientific and technological achievement matching degree ranking sequence. The scientific and technological achievements ranked at the top of the scientific and technological achievement matching degree ranking sequence have a higher degree of matching with user needs.

[0149] S4.5: Based on the value of the dynamic granularity adjustment factor, the matching degree ranking sequence of scientific and technological achievements is sorted in layers to generate a personalized list of scientific and technological achievements to be pushed.

[0150] S4.5.1: Obtain the dynamic granularity adjustment factor, compare the dynamic granularity adjustment factor with the preset granularity interval division threshold, and determine the granularity interval type to which the dynamic granularity adjustment factor belongs.

[0151] Preferably, a coarse-grained interval threshold and a fine-grained interval threshold are set, wherein the coarse-grained interval threshold represents the boundary point between coarse-grained and medium-grained division, and the fine-grained interval threshold represents the boundary point between medium-grained and fine-grained division.

[0152] Specifically, if the value of the dynamic granularity adjustment factor is less than the coarse-grained interval threshold, the user requirement is determined to be a coarse-grained requirement type; if the value of the dynamic granularity adjustment factor is greater than or equal to the coarse-grained interval threshold and less than the fine-grained interval threshold, the user requirement is determined to be a medium-grained requirement type; if the value of the dynamic granularity adjustment factor is greater than or equal to the fine-grained interval threshold, the user requirement is determined to be a fine-grained requirement type.

[0153] S4.5.2: Based on the granularity range type to which the dynamic granularity adjustment factor belongs, set the push list capacity parameter and the matching degree filtering threshold;

[0154] Furthermore, if the current user's need is a coarse-grained need, a larger push list capacity parameter and a lower matching degree filtering threshold are set to cover a wider range of scientific and technological achievements; if the current user's need is a medium-grained need, a medium push list capacity parameter and a medium matching degree filtering threshold are set to balance the push scope and accuracy; if the current user's need is a fine-grained need, a smaller push list capacity parameter and a higher matching degree filtering threshold are set to focus on highly matching scientific and technological achievements.

[0155] S4.5.3: Based on the matching degree screening threshold, perform preliminary screening on the matching degree ranking sequence of scientific and technological achievements, extract all scientific and technological achievements with a corrected multi-granularity matching degree greater than the matching degree screening threshold from the matching degree ranking sequence of scientific and technological achievements, and generate a candidate set of scientific and technological achievements to be pushed.

[0156] S4.5.4: Count the number of scientific and technological achievements in the candidate push scientific and technological achievements set, and compare the number of scientific and technological achievements with the push list capacity parameter;

[0157] Furthermore, if the number of scientific and technological achievements is less than or equal to the push list capacity parameter, then all scientific and technological achievements in the candidate push scientific and technological achievements set will be directly included in the preliminary push list; if the number of scientific and technological achievements is greater than the push list capacity parameter, then the top N scientific and technological achievements in the candidate push scientific and technological achievements set will be selected in descending order according to the corrected multi-granularity matching degree and included in the preliminary push list, where N is equal to the value of the push list capacity parameter.

[0158] S4.5.5: Based on a dynamic granularity adjustment factor, a diversity adjustment mechanism is introduced to optimize the initial push list, specifically including:

[0159] If the value of the dynamic granularity adjustment factor is less than the coarse granularity interval threshold, the diversity enhancement operation is triggered. The repetitive scientific and technological achievement groups with high technical keyword similarity are identified from the initial push list. For each repetitive scientific and technological achievement group, the scientific and technological achievement with the highest multi-granularity matching degree after correction is retained, and the remaining scientific and technological achievements are removed from the initial push list. At the same time, the different scientific and technological achievements with low technical keyword similarity to the existing scientific and technological achievements in the initial push list are selected from the candidate push scientific and technological achievement set and added to the initial push list until the number of scientific and technological achievements in the initial push list reaches the push list capacity parameter.

[0160] If the value of the dynamic granularity adjustment factor is greater than or equal to the fine-grained interval threshold, the accuracy enhancement operation is triggered. The fine-grained boundary matching degree value of each scientific and technological achievement at the fine-grained level is extracted from the initial push list, and the fine-grained matching degree contribution weight of each scientific and technological achievement is calculated.

[0161] It should be noted that the contribution weight of fine-grained matching degree is calculated based on the proportion of fine-grained boundary matching degree to the corrected multi-grained matching degree. The higher the proportion of fine-grained boundary matching degree, the greater the contribution weight of fine-grained matching degree. The scientific and technological achievements in the initial push list are rearranged in descending order according to the contribution weight of fine-grained matching degree, and the sorting position of the scientific and technological achievements in the initial push list is updated.

[0162] S4.5.6: Based on the uncertainty metric of user demand, the initial push list is labeled with confidence level;

[0163] Specifically, the uncertainty metric of user demand is obtained. If the uncertainty metric is greater than a preset uncertainty threshold, the user demand is deemed to be unclear and a low confidence label is added to each scientific and technological achievement in the initial push list. If the uncertainty metric is less than or equal to the preset uncertainty threshold, the user demand is deemed to be relatively clear and a high confidence label is added to each scientific and technological achievement in the initial push list.

[0164] S4.5.7: Encapsulate and integrate all scientific and technological achievements in the initial push list after optimization by the diversity adjustment mechanism and confidence labeling to generate a personalized scientific and technological achievement push list. The personalized scientific and technological achievement push list includes the identification information of the scientific and technological achievements, the corrected multi-granularity matching degree, the sorting position, the confidence label, and the corresponding technical keywords, application fields and achievement type attributes.

[0165] S4.5.8: For personalized science and technology achievement push lists, generate push description tags based on the granularity interval type to which the dynamic granularity adjustment factor belongs;

[0166] Furthermore, if the current user's need is coarse-grained, the push notification label is marked as a broad exploratory push, indicating that the current push results cover multiple technical directions; if the current user's need is medium-grained, the push notification label is marked as a balanced recommendation push, indicating that the current push results take into account both scope and accuracy; if the current user's need is fine-grained, the push notification label is marked as a precise matching push, indicating that the current push results focus on highly relevant results; the push notification label is attached to the personalized scientific and technological achievement push list to complete the hierarchical filtering operation based on dynamic granularity adjustment factors.

[0167] S4.6: Organize the scientific and technological achievements in the personalized scientific and technological achievement push list according to the descending order of the corrected multi-granularity matching degree, encapsulate and generate the final recommendation scheme, and output it to the user terminal interface for display.

[0168] In an optional embodiment, the platform calculates the multi-granularity matching degree between the scientific and technological achievements to be pushed and the user's needs. A new achievement, "a low-power AI chip based on asynchronous circuits," has a matching degree of 0.6 at the coarse-grained level, 0.7 at the medium-grained level, and 0.8 at the fine-grained level. After weighted fusion, a comprehensive matching degree is obtained. Then, the consistency of the matching degree at each level (smaller differences indicate higher confidence) and the uncertainty measure of user needs (0.5 indicates medium uncertainty) are combined to make corrections, resulting in a corrected matching degree of 0.72. Similar calculations are performed on all candidate achievements and they are sorted. Then, a push strategy is set according to a dynamic granularity adjustment factor of 0.6 (medium-grained demand type): achievements with a matching degree higher than 0.5 are selected to ensure coverage of a certain range while taking into account accuracy, and diversity adjustments are used to avoid the recommended results being too similar. The generated personalized push list contains 8 achievements, arranged from high to low matching degree, and each achievement is given a technical attribute description and confidence level label. The overall list is labeled as "balanced recommendation push" and output to user A's operation interface.

[0169] Example illustration, such as Figure 2 As shown, the user terminal interface displays the core indicators of user A: historical interaction frequency (18), dynamic granularity adjustment factor (0.60), demand uncertainty (0.50), and areas of interest, and uses a visual indicator bar to show the position of the granularity factor in the coarse-medium-fine granularity range.

[0170] In summary, this invention introduces a dynamic granularity adjustment factor to partition the dataset of scientific and technological achievements and construct a multi-dimensional feature vector space. This allows the features of scientific and technological achievements and the features of users' historical behavior to be adaptively expressed at different scales, avoiding information redundancy or feature loss caused by fixed granularity. Based on this, an improved multi-granularity rough set model with an adaptive boundary adjustment mechanism is constructed to dynamically characterize the decision boundaries at different granularity levels, enhancing the model's ability to discriminate complex data distributions and uncertain samples. By calculating the upper and lower approximate sets of users' historical behavior in the multi-granularity rough set model, a fuzzy boundary region of user needs is generated, achieving explicit modeling of users' implicit needs and needs uncertainty. Based on the fuzzy boundary of scientific and technological achievements features and user needs, the corrected multi-granularity matching degree is calculated and a personalized push list is generated, ensuring that the recommendation results reflect both the degree of needs matching and the stability of needs. This improves the overall accuracy, adaptability, and practical application value of scientific and technological achievements push.

[0171] This embodiment also provides a computer device applicable to the technology achievement promotion method based on improved multi-granularity rough sets, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the technology achievement promotion method based on improved multi-granularity rough sets proposed in the above embodiment.

[0172] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0173] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for pushing scientific and technological achievements based on an improved multi-granularity rough set, as proposed in the above embodiments.

[0174] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0175] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for promoting scientific and technological achievements based on improved multi-granularity rough sets, characterized in that: include, A dataset of scientific and technological achievements is obtained, and the dataset is divided into granularities by introducing a dynamic granularity adjustment factor to construct a multidimensional feature vector space. The dataset of scientific and technological achievements includes a set of user historical behavior data and a set of scientific and technological achievement feature data. Based on the multidimensional feature vector space, an improved multi-granularity rough set model is constructed, wherein the improved multi-granularity rough set model dynamically adjusts the boundary region range of different granularity levels by introducing an adaptive boundary adjustment mechanism. Calculate the upper and lower approximation sets of each element in the user historical behavior data set in the improved multi-granularity rough set model to generate a fuzzy boundary region for user needs. Based on the set of feature data of scientific and technological achievements and the fuzzy boundary region of user needs, the multi-granularity matching degree between scientific and technological achievements and user needs is calculated, and a personalized scientific and technological achievement push list is generated through the granularity division result, and the final recommendation scheme is output. The method for constructing the dynamic granularity adjustment factor is as follows: The user historical behavior data set is structured and multiple behavioral feature fields are extracted from it. Attribute parsing is performed on the set of scientific and technological achievement feature data to extract multiple feature attribute fields from the set of scientific and technological achievement feature data; Based on the aforementioned multiple behavioral feature fields, the number of times each user interacts with various scientific and technological achievements within a preset time window is counted, and the user's historical interaction frequency is calculated. Based on the aforementioned multi-class feature attribute fields, a similarity algorithm is used to calculate the similarity of achievement features between any two scientific and technological achievements; Based on the frequency of user historical interactions and the similarity of the results features, a dynamic granularity adjustment factor is constructed through a nonlinear mapping function. The method for constructing the improved multi-granularity rough set model is as follows: Based on a multidimensional feature vector space, a decision information system is constructed by taking all feature vector points in the multidimensional feature vector space as the universe of discourse and the attributes of each dimension as the set of conditional attributes. A roughness threshold is set based on the number of attributes and the distribution of attribute values ​​in the conditional attribute set. At the same time, a fine-grained weight coefficient is set based on the distinguishing ability of each attribute in the classification decision. Based on the set of conditional attributes and the fine-grained weight coefficients, a weighted indistinguishable relation of multiple granular levels is constructed on the universe of discourse, wherein the multiple granular levels include coarse-grained level, medium-grained level and fine-grained level; The universe of discourse is divided according to the weighted indistinguishable relation to generate a set of equivalence classes at this granularity level; An adaptive boundary adjustment mechanism is introduced, wherein the adaptive boundary adjustment mechanism dynamically calculates the boundary adjustment factor based on the roughness threshold and the distribution density of the equivalence class set; Based on the boundary adjustment factor, the adjusted boundary region range is calculated for the multiple granularity levels, and the dynamic boundary region corresponding to each granularity level is generated. The decision information system, the roughness threshold, the fine-grained weight coefficient, the weighted indistinguishable relation, the adaptive boundary adjustment mechanism, and the dynamic boundary region are integrated and encapsulated to construct an improved multi-granularity rough set model.

2. The method for promoting scientific and technological achievements based on improved multi-granularity rough sets as described in claim 1, characterized in that: The method for outputting the final recommendation scheme is as follows: The corrected multi-granularity matching degrees are sorted in descending order of numerical value to generate a scientific and technological achievement matching degree ranking sequence. Based on the value of the dynamic granularity adjustment factor, the matching degree ranking sequence of the scientific and technological achievements is hierarchically filtered to generate a personalized scientific and technological achievement push list. The scientific and technological achievements in the personalized scientific and technological achievement push list are organized in descending order according to the corrected multi-granularity matching degree, packaged to generate the final recommendation scheme, and output to the user terminal interface for display.

3. The method for promoting scientific and technological achievements based on improved multi-granularity rough sets as described in claim 2, characterized in that: The method for calculating the corrected multi-granularity matching degree is as follows: Extract scientific and technological achievement objects one by one from the set of scientific and technological achievement feature data, and vectorize the technical keywords, application fields and achievement types of each scientific and technological achievement object to generate a set of scientific and technological achievement feature vectors; Based on multiple granularity levels, the granularity matching degree between the feature vector of the scientific and technological achievement and the fuzzy boundary region of user needs is calculated according to the corresponding equivalence class set and the adjusted boundary region, and multi-granularity matching degree is generated. Based on the uncertainty metric of user needs, the confidence level of the multi-granularity matching degree is corrected to generate the corrected multi-granularity matching degree.

4. The method for promoting scientific and technological achievements based on improved multi-granularity rough sets as described in claim 3, characterized in that: The method for generating the fuzzy boundary region of the user requirement is as follows: Elements are extracted one by one from the user historical behavior data set and mapped to the domain of the improved multi-granularity rough set model to generate a set of user behavior objects to be analyzed. For each object in the set of user behavior objects to be analyzed, calculate the approximate set and the upper approximate set of the multiple granularity levels, and calculate the difference between the upper approximate set and the lower approximate set to generate the original boundary region corresponding to each granularity level; Based on the adaptive boundary adjustment mechanism and boundary adjustment factor, the original boundary regions at each granularity level are respectively corrected to generate the adjusted boundary regions corresponding to each granularity level. Based on the adjusted boundary region, a fuzzy membership quantification method is introduced to calculate the degree of membership of each object to the target concept. Based on the fuzzy membership value, a multi-granularity fusion strategy is used to comprehensively integrate the boundary information of the multiple granularity levels to generate the fuzzy boundary region required by the user. Calculate the uncertainty metric of user needs based on the fuzzy boundary region of the user needs.

5. The method for promoting scientific and technological achievements based on improved multi-granularity rough sets as described in claim 1, characterized in that: The method for constructing the multidimensional feature vector space is as follows: Obtain a dynamic granularity adjustment factor, and compare the dynamic granularity adjustment factor with a preset granularity threshold to obtain a set of scientific and technological achievement features; When the value of the dynamic granularity adjustment factor is greater than the preset granularity threshold, a fine-grained partitioning strategy is used to perform feature decomposition on the scientific and technological achievement dataset to generate a fine-grained set of scientific and technological achievement features. When the value of the dynamic granularity adjustment factor is less than or equal to the preset granularity threshold, a coarse-grained partitioning strategy is used to aggregate features of the scientific and technological achievement dataset to generate a coarse-grained scientific and technological achievement feature set. The set of features of the scientific and technological achievements is vectorized and encoded to construct a multi-dimensional feature vector space.

6. The method for promoting scientific and technological achievements based on improved multi-granularity rough sets as described in claim 1, characterized in that: The user historical behavior data set includes browsing history, download history, and tags of areas of interest; the scientific and technological achievement feature data set includes technical keywords, application fields, and achievement types; the dynamic granularity adjustment factor is adaptively adjusted according to the frequency of user historical interactions and the similarity of achievement features.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the technology achievement push method based on any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the technology achievement push method based on any one of claims 1 to 6.

Images