Academic hotspot-driven economic return prediction method

Abstract

The invention discloses an academic hotspot-driven economic return prediction method, which relates to the technical field of scientific research analysis, and aims at the data of a multi-source academic social platform, the method comprises the following steps of: removing redundant posts and filtering noise by using a multi-dimensional similarity function, identifying academic entities and constructing a fusion type semantic vector based on a BERT model and an academic dictionary, and predicting the economic return based on the fusion type semantic vector. Through multi-layer clustering and time sequence activeness analysis, early hot spots are mined, time-delay correlation integral and potential influence scoring are carried out, the association degree and transformation potential of the early hot spots and official scientific research achievements are evaluated, finally, a cross-department collaboration degree index and a technical gap index are combined, high-potential topics are automatically pushed, patent layout is optimized, and the method is suitable for popularization and application. According to the method, closed-loop management from leading-edge signal capture to achievement incubation is realized, and the efficiency and accuracy of colleges and universities in the aspects of capturing interdisciplinary opportunities, coordinating resource allocation and promoting patent transformation are greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of scientific research analysis technology, and in particular to an economic return prediction method driven by academic hotspots. Background Art

[0002] As the global research ecosystem accelerates its digitization and networking, the channels for generating and disseminating academic information have expanded from traditional academic journals, patent databases, and research project archives to encompass a multi-source academic social networking platform. Researchers from universities and research institutions often instantly share preprint links, meeting notes, technical roadmaps, and even real-time experimental data on these platforms, allowing "early academic signals" to spread in seconds. Simultaneously, decisions regarding scientific and technological funding, patent portfolio development, and industry incubation increasingly rely on the rapid capture and valuation of these early signals: managers seek to understand the potential for subsequent research output, potential technological gaps, and cross-departmental collaboration opportunities while hot topics are still in their infancy, enabling targeted investment and patent preemption. However, the vast majority of posts are filled with irrelevant entertainment content, marketing messages, cross-platform reposts, and distorted text. Furthermore, objective differences across platforms, such as heterogeneous data structures, varying timestamp granularity, and a lack of user tags, significantly dilute "valuable research dynamics" with noise. If the school still relies solely on formal literature databases or manual searches, it may miss the strategic layout window at the turning point of key disciplinary intersections, resulting in high-value cutting-edge patents and industrial opportunities being preempted by others, and even a serious mismatch between scientific research funding and economic returns.

[0003] In the above application scenarios, the core technical problems that need to be solved urgently are:

[0004] How to achieve millimeter-level precision deduplication and anomaly filtering in high-noise, multi-source heterogeneous academic social data streams, and extract early academic hotspots in real time based on deep semantic analysis and time series modeling, and then perform time-lag correlation mapping with formal scientific research results libraries to quantitatively evaluate their potential economic returns and patent conversion value.

[0005] If this problem is not properly resolved, managers will face three major consequences: First, misjudgment or omission of hotspot identification will directly lead to deviation in the direction of scientific research funding and imbalance in team resource allocation; second, the lack of accurate characterization of the hotspot-result lag will cause the best window for patent application to be missed, reducing patent coverage and magnifying technological gaps; third, the inability to quantify the relationship between hotspots and economic benefits will weaken the timeliness of cross-departmental collaboration and industry-university-research docking mechanisms, ultimately affecting the competitive advantages of universities and enterprises in the cutting-edge technology track.

[0006] To this end, the present invention provides an economic return prediction method driven by academic hotspots. Summary of the Invention

[0007] (1) Technical problems solved

[0008] In response to the shortcomings of the existing technology, the present invention provides an economic return prediction method driven by academic hotspots. By targeting data from multi-source academic social platforms, a multi-dimensional similarity function is first used to remove redundant posts and filter noise. Then, based on the BERT model and academic dictionary, academic entities are identified and fused semantic vectors are constructed. Through multi-layer clustering and time series activity analysis, early hotspots are mined and time-lag correlation integrals and potential influence scores are performed to evaluate their correlation with formal scientific research results and their transformation potential. Finally, combined with the cross-departmental synergy index and technology gap index, high-potential topics are automatically pushed and patent layout is optimized, realizing closed-loop management from cutting-edge signal capture to results incubation. This method greatly improves the efficiency and accuracy of colleges and universities in capturing interdisciplinary opportunities, collaborative resource allocation, and promoting patent transformation; thereby solving the technical problems recorded in the background technology.

[0009] (2) Technical solution

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0011] The economic return prediction method driven by academic hot spots includes: capturing original posts from multiple academic social platforms, determining duplicate content through a multi-dimensional similarity function, merging or deleting content if the MDI value exceeds a certain threshold, performing anomaly detection and noise filtering on the remaining text, and outputting a structured preliminary cleansed dataset;

[0012] Based on the pre-trained BERT model and customized academic dictionary, we perform academic entity recognition and multi-dimensional text judgment on each cleaned data item, remove irrelevant information, and perform fused semantic vector mapping on the retained academic text to obtain a collection of academic vectorized texts.

[0013] At the global level of the academic vectorized text collection, we extract major categories, subdivide these categories into subcategories, and generate preliminary cluster labels. We then perform activity statistics on topics over a time window and detect sudden increases using a cumulative increment function. If the activity increment threshold and academic value threshold are met, we output the corresponding early hotspot.

[0014] We map early hot topics with existing formal scientific research results in a two-way manner, and use the time-lag correlation integral to measure the linkage between topics and scientific research outputs after the peak of popularity, combined with the potential influence score Π k , measure its translation potential and finally present the time-lagged evolution and high-impact topics in a visual way;

[0015] Based on hot topic information, the cross-departmental synergy index is calculated to identify the best teams or departments, assist in dynamic push and incentive measures, and measure potential patent coverage with the technology gap index. If the technology gap index is higher than the preset gap threshold, an external prompt will be issued.

[0016] Furthermore, according to the predefined target academic topics or keywords, corresponding posts are captured from the public platform and recorded as {P i,j} and with basic meta-information, introduce the multi-dimensional similarity function MDI(P i,j ,P i,k );

[0017] If the obtained multidimensional similarity function MDI value is higher than the preset threshold, redundant data will be deleted and the only valid version will be retained, and the remaining posts will be mapped to fields in a unified format; abnormalities will be filtered out in combination with the noise characteristics of the academic social platform, including: checking for extremely abnormal post lengths, full symbols or full links and other noise patterns, verifying texts that were not fully parsed during the crawling process, and eliminating unreadable or severely distorted content.

[0018] Furthermore, the post texts are parsed one by one, where the pre-trained BERT model is combined with a customized academic dictionary to perform academic entity recognition, marking features such as research directions, technical terms, and author citations as {E k}, for each post P s Generate word segmentation sequence T s , record word units and tagged academic entity location information;

[0019] Perform multi-dimensional text judgment on each post, using the judgment function Ω(P s ), if Ω(P s ) returns a true value, the post is retained and output to the post collection; otherwise, it is marked as entertainment content and removed.

[0020] Furthermore, the same pre-trained BERT model is called to obtain the basis vector V base (P s ), integrating academic entity weights and contextual relationships to obtain the complete vector V(P s ), defined as follows:

[0021]

[0022] Where: V base (P s ) is the model for the post P s The global embedding of E k (P s ) represents the semantic position of the kth academic entity in the post, Φ(·) is the local context embedding function; γ k is the fusion weight.

[0023] Furthermore, for each post P s The embedding representation V(P s ) performs multi-level clustering, where a vector matrix is constructed Where n is the number of posts, m is the vector dimension, and each row corresponds to a post vector V(P s ); use pre-selected hierarchical clustering at the global level to extract several major categories, and further subdivide each major category based on the distribution of academic entities, and subdivide similar professional directions into more precise topic subcategories.

[0024] Furthermore, we perform time series statistics on each topic, record the discussion volume and interaction indicators of posts in discrete time periods, and record the comprehensive activity vector of a topic k at time τ as h k (τ), in order to detect short-term spikes, the following formula is defined as the cumulative increment function Λ of the multidimensional activity curve k (t), if the cumulative increment function Λ k (t) If a sudden jump occurs within a certain time interval, the corresponding topic is determined to be a potential early hot spot;

[0025]

[0026] Among them, h k (τ) represents the activity vector of topic k at time r, Its gradient at τ, w k is the weight vector corresponding to topic k, and <·,·> is the inner product operation.

[0027] Furthermore, according to the cumulative increment function Λ k (t) The maximum increment interval of the post is used to filter the post text of the corresponding sub-topic, and the sub-topics that pass the screening are marked as early signals. For topics that do not meet the threshold or have a sudden increase but low content value, they are marked as ordinary discussions; each early hot topic and its activity curve are extracted, and the key academic entities and tags of each hot topic are searched in the formal scientific research database to obtain the corresponding research results entries.

[0028] Furthermore, in order to evaluate the time lag between the appearance of hot topics on social platforms and the subsequent output of corresponding formal results, and to comprehensively score the potential influence, the following time lag correlation score γ is defined: k (Δ) is used to measure the degree of linkage between the two. The formula is as follows:

[0029]

[0030] Where: γ k (Δ) is the time offset of topic k at Δ, z k (τ) represents the social media activity vector of topic k at time τ; r k(σ) represents the formal scientific research achievement record corresponding to topic k at time σ; Φ(·) is the high-level mapping of the formal scientific research achievement vector; <·,·> is the inner product operation, exp(α·<·>) is the exponential kernel function, α is the balance coefficient; δ(·) is the Dirac function, and [t0,t1] is the time observation interval.

[0031] Furthermore, the potential influence score Π is introduced k To quantify the potential impact of hot topics on formal scientific research results under different time lag conditions, the following weighted cumulative integral form can be designed:

[0032]

[0033] Where: k is the comprehensive potential influence score of hot topic k, γ k (Δ) is the double integral relevance of the topic relative to the formal scientific research output, ψ(·) is the enhancement or suppression function; κ(Δ) is the time lag weighting function, Δ max is the maximum observation lag.

[0034] Furthermore, a potential influence score Π is generated k and the peak lag Δ * Then, multi-dimensional visualization and key topic prompts are performed, in which the activity curve of each topic (k) in social media, the publication or citation curve of the corresponding scientific research results, and Δ * The significant linkage points at Δ are projected into the same coordinate system to form a time superposition diagram of influence-time lag scatter diagram: * is the horizontal axis, Π k As the vertical axis, the hot topics are scatter-visualized, and the color or size can represent different academic fields or the overlap with existing literature;

[0035] If Π k Exceeding the scoring threshold ω impact , and its Δ * Below the critical value ω delay , marked as rapid conversion type; has not yet appeared in formal literature, but Π k Topics that are significantly higher than the average level in the same field are marked as potentially explosive and will be highlighted or prompted to university administrators or research teams in the visual interface through pop-up windows.

[0036] Furthermore, combined with the time-delayed correlation integral γ k (Δ) and potential influence score Π k , hot topics are presented in a multi-dimensional way in a visual interface, where the topic activity curve and the cumulative release curve of the corresponding formal scientific research results are displayed on the same timeline, showing the lagging relationship between the two;

[0037] The peak value of the time delay of each topic Δ * Together with the influence score, it is projected into a two-dimensional or three-dimensional coordinate system to focus on identifying cutting-edge topics with short lag periods and high growth potential; hot topics that have not yet appeared in a large number of formal literature but have continued to increase in discussion on social platforms and whose time-lag analysis shows high potential are marked significantly.

[0038] Furthermore, we read the hot topics marked as potential explosive or rapid transformation, and based on their academic field labels, time-lag impact measure γ k (Δ) and potential influence score Θ information to construct the cross-departmental synergy index Θ col (k,a), through the cross-departmental synergy index Θ col (k,a) selects the departments or teams with the highest coupling degree with topic k, and uses the cross-departmental synergy index Θ col The ranking of the most promising topics will be pushed to the heads of departments or relevant disciplines; among them,

[0039]

[0040] Among them, S k (τ) represents the academic entity distribution vector of topic k at normalized time τ, M a (σ) represents the research capability vector of department a at normalized time σ; Φ(·) is the mapping function, <·,·> is the inner product operation, exp(β·<·,·>) is the exponential kernel function, β is the balance coefficient; κ(τ,σ) is the time weight kernel function.

[0041] Furthermore, for the hot topic k, we search for existing patents or patent application documents in the same field and calculate the technology gap index Ω gap (k) Used to judge patent coverage density and innovation space, as follows:

[0042]

[0043] Where: E k Represents the multidimensional concept vector of topic k, is the concept vector of the i-th patent document, is the distance measurement function; construct the corresponding demand vector D for the external subject v , combined with S k (t) performs inner product matching on the academic entity distribution to identify highly compatible industrial partners.

[0044] (3) Beneficial effects

[0045] The present invention provides an economic return prediction method driven by academic hot spots, which has the following beneficial effects:

[0046] 1. Accurately mine key entities such as research topics and technical terms within posts. This generates a word segmentation sequence for each post and records the entity's location within the text, providing a basis for in-depth semantic analysis. Entity tagging allows subsequent vectorization to better distinguish between specialized terms and common words, improving the accuracy of clustering and hotspot discovery.

[0047] 2. Using indicators such as academic entity distribution and the proportion of core words in texts, we can accurately filter out entertainment or advertising posts and retain only posts with high academic weight, so that subsequent vectorization and clustering can focus more on real scientific research discussions and ensure the concentration of academic value: Using the integral kernel function Ω(P s ) and other methods to achieve more flexible judgment of academic tendencies, not limited to a single rule, and reduce the risk of accidental deletion.

[0048] 3. Track the activity of topics in daily / weekly / monthly dimensions to promptly identify possible explosive discussions; multiple dimensions such as the number of posts, interaction volume, and keyword frequency can be introduced to use Λ k (t) or gradient detection algorithm accurately locates the period of sudden increase, and only marks topics with significant and sustained jumps as potential early hot spots, reducing the risk of misjudgment caused by instantaneous hype and preventing short-term noise.

[0049] 4. By combining multi-layer clustering with time series analysis, we can quickly identify research topics with high potential or innovation within the vast academic social media landscape. Our progressive screening strategy ensures the accuracy and interpretability of our findings, providing a solid foundation for subsequent comparisons with official research findings and visualization.

[0050] 5. Comparing the topic activity curve and the official achievement release curve in the same coordinate system can help managers intuitively understand the time lag relationship. * and potential influence score Π k Using this as a coordinate, we can quickly identify hot topics with low latency and high impact. This highlights potential explosive topics that have not yet appeared widely in formal literature but have a clear upward trend in academic research, allowing managers to intervene and provide support in advance.

[0051] 6. By calculating the cross-departmental synergy index Θ col (k, a), automatically find the department or team that best matches the topic, which can achieve accurate push. It is recommended to set up interdisciplinary research incentives, accelerate the formation of cross-domain joint teams, improve the response speed and R&D efficiency to early hot spots, encourage interdisciplinary resource integration, combine the potential value of topics, departmental research capabilities and management decisions, form an efficient mechanism for flexible allocation of internal resources, and realize closed-loop linkage.

[0052] 7. Through the technical gap index Ω gap(k) Measure the coverage of topics and existing patents, identify areas with room for innovation, and facilitate the connection between industry, academia, and research: introduce the demand vectors of industrial partners, match them with academic entities, quickly lock in cooperation opportunities, incubate technological achievements, and provide all-round support for their implementation in actual application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of the economic return prediction method driven by the academic hotspot of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0055] See also Figure 1 The present invention provides an economic return prediction method driven by academic hot spots, including:

[0056] Step 1: Capture original posts on multi-source academic social platforms i,j}, determine duplicate content through a multidimensional similarity function, merge or delete the remaining content if the MDI value exceeds the corresponding threshold, perform anomaly detection and noise filtering on the remaining text, and output a structured preliminary cleansing data set;

[0057] The step 1 includes the following:

[0058] Step 101: According to the predefined target academic topic or keyword, corresponding posts are captured from the public platform, which are recorded as {P i,j}, where P i,j It represents the jth original post on platform i, along with its publisher ID and timestamp and other basic metadata, which are uniformly stored in dataset D 101 ; in D 101 Based on the introduction of multidimensional similarity function MDI (P i,j ,P i,k ) to determine the degree of duplication between the two posts, as follows:

[0059]

[0060] Where: R is the total number of embedding vector dimensions; E r (P i,j ) represents the r-th dimension embedding vector for the post P i,j, cos(·,·) is used to calculate the cosine similarity of two vectors in the r-th dimension embedding space; β is the balance coefficient, which is used to adjust the contribution of similarity to the exponential function;

[0061] If the obtained multidimensional similarity function MDI(P i,j ,P i,k ) value is higher than the preset threshold, it can be determined that the two posts are duplicates or highly similar, and redundant data is deleted and the only valid version is retained. After the deduplication is completed, the remaining posts are mapped to a unified format (including text content, platform source, timestamp, user ID), and the normalized D 102 Dataset;

[0062] When in use, original posts are obtained from multiple academic social platforms to ensure data diversity and completeness, improve data coverage, and determine duplicate and highly identical content based on the multidimensional similarity function MDI. This can effectively reduce the subsequent analysis distortion caused by duplicate posts and prevent repeated interference. After formatting and mapping, all posts are stored in a standardized structure to facilitate cross-platform data integration and alignment. Noise features such as extreme length, full symbols or full links are eliminated to ensure the overall quality of the data. Records that cannot be parsed or whose content is severely distorted during the crawling process are cleaned up to avoid occupying subsequent analysis resources. During the cleaning process, the post similarity information is marked accordingly to facilitate secondary comparison or data backtracking in the future.

[0063] Step 102: Combine the noise characteristics of the academic social platform to 102 Perform anomaly filtering, including checking for abnormal post lengths, noise patterns such as full symbols or full links, checking for text that was not fully parsed during the crawling process, and removing unreadable or severely distorted content. After removing abnormal posts, organize the valid text and its basic identifiers into structured records. 103 , and retain the similarity mark in the MDI calculation process so that relevant records can be referenced for secondary detection later;

[0064] When used, after multiple filtering steps, it can output highly credible and readable text records, retaining only posts with genuine academic background or potential research value, improving the accuracy of subsequent NLP processes and laying the foundation for semantic analysis:

[0065] Effectively blocking invalid posts, advertisements, and other noise at this stage can reduce the computational burden of subsequent analysis. This step utilizes multi-dimensional similarity determination, anomaly filtering, and structured output to form high-quality preliminary cleansing data, laying a solid foundation for subsequent semantic parsing, topic clustering, and time series analysis. At the same time, standardization is performed in terms of data format, deduplication, and tagging, greatly improving the efficiency and reliability of subsequent processing.

[0066] Step 2: Based on the pre-trained BERT model and customized academic dictionary, perform academic entity recognition and multi-dimensional text judgment on each cleaned data item, remove irrelevant information, and perform fused semantic vector mapping on the retained academic text to obtain a collection of academic vectorized texts;

[0067] The second step includes the following:

[0068] Step 201: Extract structured record D 103 , parsing the post text one by one, using the pre-trained BERT model combined with a customized academic dictionary to perform academic entity recognition, marking features such as scientific research direction, technical terms, and author citations as {E k}, for each post P s Generate word segmentation sequence T s , record word units and marked academic entity location information, and store these entity tags in the data object D 201 ;

[0069] Using pre-trained BERT and a customized academic dictionary, we can accurately mine key entities such as research topics and technical terms within posts, generate word segmentation sequences for each post, and record the entity's location within the text, providing a basis for in-depth semantic analysis. Entity tagging allows subsequent vectorization to better distinguish between specialized terms and common words, improving the accuracy of clustering and hotspot discovery.

[0070] Step 202: Based on the identified academic entity distribution and post context, 201 In the multi-dimensional text judgment, each post is judged, and posts that mainly contain traces of entertainment, life or commercial promotion are eliminated. The following judgment function Ω(P s ):

[0071]

[0072] Where: P s represents the sth academic social post, δ(E k (P s ))Representation post P s The total weight or comprehensive score of academic entities (such as academic terms, research methods, etc.): θ(P s ) captures the external statistical features of the post text (such as the proportion of core words, external link density, text structure, etc.); Ψ(δ(E k (P s )),θ(P s),t) is a vector function that varies with the parameter t∈[0,1], which is used to integrate academic entity information and text features, and can implement polynomial or exponential mapping internally to more flexibly measure the academic tendency of the content. M is a pre-set weighting matrix used to highlight dimensions that are more sensitive to academic elements. <·,·> represents the inner product operation or other linear mapping between the matrix and the vector; Γ(·) is a judgment operator that can implement a threshold or step function, which is used to finally output whether the post meets the academic retention conditions;

[0073] In this formula, by changing δ(E k (P s ) and θ(P s ) performs matrix weighting and integrates it over the interval [0, 1] to capture the distribution and fluctuation of the academic content of the post in a wider feature space. Finally, Γ(·) determines whether the post passes the academic judgment based on the inner product result. This can effectively avoid over-reliance on a single indicator, improve the accuracy of eliminating entertainment or promotional posts, and provide purer academic text data for subsequent natural language processing and hot spot mining steps.

[0074] If Ω(P s ) returns a true value, then the post is retained and output to the post set D 202 Otherwise, it is marked as entertainment content and removed to ensure that subsequent vectorization focuses on text with high academic value;

[0075] When used, multi-dimensional content judgment: using indicators such as academic entity distribution and the proportion of core words in the text, accurately screen out entertainment or advertising posts, and only retain posts with high academic weight, so that subsequent vectorization and clustering are more focused on real scientific research discussions, which can ensure the concentration of academic value: using the integral kernel function Ω(P s ) and other methods to achieve more flexible judgment of academic tendencies, not limited to a single rule, and reduce the risk of accidental deletion.

[0076] Step 203: The post collection D retained in step 202 202 , call the same pre-trained BERT model to get the basis vector V base (P s ), and then integrate the academic entity weight and contextual relationship to obtain the complete vector V(P s ), defined as follows:

[0077]

[0078] Where: V base (P s ) is the model for the post P s The global embedding of E k (P s) represents the semantic position of the kth academic entity in the post, Φ(·) is the local context embedding function, which is used to extract the semantic information of the entity’s surrounding words; γ k The fusion weight is used to amplify the core topic features related to the entity and to convert all the posts corresponding to V(P s ) vector, stored as an academic vectorized text collection D 203 , and retain academic entity mapping information in the same data structure so that early signal mining and emerging hotspot identification can seamlessly call these vectors and entity labels;

[0079] Through academic entity detection, multi-dimensional determination of academic posts, and fusion semantic vector mapping, a collection of academic vectorized texts D that can be used for subsequent clustering or topic modeling is formed. 203 ;

[0080] When used, integrate global and local semantics: through V base (P s ) and academic entity weight γ k Φ(E k (P s )) integration, taking into account both general context and professional domain knowledge, high-value entities gain greater weight in the vector space and are more easily identified as potential hotspots by subsequent algorithms, which can enhance the influence of key terms.

[0081] By eliminating irrelevant content through multi-dimensional analysis and expressing academic posts as high-dimensional vectors, step two significantly improves the data's fidelity to academic domain characteristics, enabling subsequent early-stage hotspot discovery to be conducted in a more precise semantic space. Furthermore, combining entity annotation with semantic fusion provides flexible and scalable technical support for identifying interdisciplinary disciplines and emerging research directions.

[0082] Step 3: Extract major categories from the global level of the academic vectorized text collection, subdivide subcategories within each major category, and generate preliminary cluster labels. Statistically analyze the activity of topics over a time window and detect sudden increases using a cumulative increment function. If the activity increment threshold and academic value threshold are met, the corresponding early hot spots are output.

[0083] The step three includes the following:

[0084] Step 301: Based on the academic vectorized text set D 203 , for each post P s The embedding representation V(P s ) performs multi-level clustering to identify potential academic directions as follows:

[0085] Constructing a vector matrix Where n is the number of posts, m is the vector dimension, and each row corresponds to a post vector V(P s); at the global level, a pre-selected hierarchical clustering is used to extract several major categories (such as intelligent manufacturing, genomics, etc.) and obtain preliminary clustering labels D 301 ; Further subdivide each major category according to the distribution of academic entities, subdivide similar professional directions into more precise topic subcategories, and save this subcategory tag in the preliminary clustering tag D 301 , so that the evolution of each topic can be tracked more finely in subsequent time series analysis;

[0086] When using this method, first obtain the macro-domain (large category), then subdivide the large category into subcategories to improve the clarity and interpretability of the topic structure. Combining the clustering results with the entity labels obtained in step 2, combined with the academic entity distribution, can more accurately distinguish similar directions and track subsequent evolution in detail. Preliminary cluster labeling provides a topic granularity foundation for subsequent time series activity analysis and topic surge detection.

[0087] Step 302: After completing the sub-topic division, the preliminary clustering mark D 301 Perform time series statistics on each topic in the post, record the discussion volume and interaction indicators of the post in a discrete time period (such as day, week, month), and record the comprehensive activity vector of a topic k at time τ as h k (τ);

[0088] To detect short-term spikes, the following formula is defined as the cumulative increment function Λ of the multidimensional activity curve: k (t):

[0089]

[0090] If the cumulative increment function Λ k (t) A sudden jump occurs within a certain time interval, indicating that the topic has heated up rapidly in a short period of time and can be determined as a potential early hot spot. After the detection is completed, the time series curve of each topic and the detection results are stored together as the time series activity and sudden increase detection result D 302 , and indicate the specific time period when the sudden increase occurred;

[0091] When in use, dynamically capture discussion heat: track the activity of topics in dimensions such as day / week / month, and promptly discover possible explosive discussions; multi-dimensional indicator integration: multiple dimensions such as the number of posts, interaction volume, and keyword frequency can be introduced to use Λ k (t) or gradient detection algorithm accurately locates the period of sudden increase, and only marks topics with significant and sustained jumps as potential early hot spots, reducing the risk of misjudgment caused by instantaneous hype and preventing short-term noise.

[0092] Step 303: After obtaining the sudden increase period of each topic, comprehensive academic entity analysis and cluster weights are used to conduct a secondary screening of discussion points that meet the sudden increase threshold and have research foresight, as follows:

[0093] According to Λ k (t) The maximum increment interval, screen the post text of the corresponding sub-topic, confirm its originality or potential academic value in the academic field, mark the sub-topic that passes the screening as an early signal, and write it into the emerging hotspot screening and output together with its activity curve, detection time period and other information. 303 For topics that do not meet the threshold or have a sudden increase but low content value, they are marked as ordinary discussions for later management or review; on the basis of retaining the high-dimensional semantic information of academic posts, through multi-layer clustering and time series sudden increase detection, emerging topics that have not yet appeared in formal literature on a large scale are effectively identified, and an activity change curve is generated for each topic, which is written into the emerging hot spot screening and output D 303 , including early hot spots and their sudden increase periods, cluster labels and activity gradient information;

[0094] It should be noted that the activity increment threshold is calculated by Λ k (t) describes the overall incremental change of topic k. If Λ k (t) The change in the specified time window exceeds the preset τ jump Or if there is a significant jump in its change curve (such as a growth rate exceeding a certain proportion relative to the previous cycle), it can be regarded as reaching the activity increment threshold. This threshold can be dynamically adjusted according to the distribution of historical data to ensure that only topics that have truly seen a significant increase in discussion heat are marked.

[0095] The threshold for judging academic value is to determine the originality or potential academic value of posts in the corresponding subtopic by comprehensively considering the density of academic terms, the number of citations of credible literature, and the professional background of the discussants, so as to set one or more minimum scores / ratios τ. academic , if the overall academic score or keyword distribution of posts within the sub-topic does not reach |τ academic Even if its discussion volume surges (high activity), it can still be regarded as an early research hotspot with low content value and not listed as a priority;

[0096] Some subtopics may be excluded even though they have a surge in popularity within a short period of time, but the total number of posts or duration is short. In this case, a minimum number of posts τ can be set. size or the shortest duration τ duration , to avoid false spikes caused by instantaneous water stickers;

[0097] During use, multiple thresholds, such as academic value judgment thresholds and activity increment thresholds, are used to eliminate explosive topics lacking depth. A secondary screening process ensures quality, retaining only those with sufficient academic weight or innovation that have seen sudden increases, focusing on truly high-value potential: forming a list of "early hot spots." Control conditions, such as the minimum number of posts and duration of postings, further filter out frothy discussions, improve the stability of hot spot identification, and avoid false spikes.

[0098] By combining multi-layer clustering with time series analysis, Step 3 effectively identified research topics with high potential or innovation within the vast academic social media landscape. This progressive screening strategy ensured the accuracy and interpretability of the results, providing a solid data foundation for subsequent comparisons with official research findings and visualization.

[0099] Step 4: Map the early hot spots to the existing formal scientific research results in a two-way manner, and use the time-delayed correlation integral γ k (Δ) measures the linkage between a topic and scientific research output after its peak, combined with the potential influence score Π k , measure its translation potential and finally present the time-lagged evolution and high-impact topics in a visual way;

[0100] The step 4 includes the following contents:

[0101] Step 401: Read the output emerging hot spots, filter them, and output D 303 , extract each early hot topic and its activity curve, search the official scientific research database (such as paper library, project library) for the key academic entities and tags of each hot topic to obtain the corresponding research results entries (such as title, author, publication date), and organize the matched results records into formal scientific research results mapping and time series matching D 401 and bidirectionally map them with the corresponding topics to ensure that subsequent timeline comparison analysis can be performed based on the same academic tags;

[0102] When used, key academic entities or topic tags are used to achieve consistency with official scientific research results, facilitating time series analysis and subsequent visualization. This builds a foundation for time lag research: clarifying the temporal relationship between the emergence of social hot spots and the publication of research results, preparing data alignment for time lag-related integration.

[0103] Step 402: After obtaining the formal scientific research results mapping and time series matching D 401 Finally, in order to evaluate the time lag relationship between the appearance of hot topics on social platforms and the subsequent output of corresponding formal results, as well as to comprehensively score the potential influence, the following time lag correlation score γ is defined k (Δ) is used to measure the degree of linkage between the two. The formula is as follows:

[0104]

[0105] Where: γ k (Δ) is the overall correlation between topic k and the corresponding formal scientific research results at the time offset Δ, z k (τ) represents the social media activity vector of topic k at time τ (which may include multiple dimensions such as the number of posts and keyword density);

[0106] r k (σ) represents the formal scientific research results record corresponding to topic k at time σ (such as the number of papers published, the number of citations, the progress of projects, etc.), which is also a multidimensional vector; φ(·) is the advanced mapping of the formal scientific research results vector (such as logarithmic transformation, nonlinear feature extraction), which is convenient for comparison with z k (τ) performs semantic or order-of-magnitude alignment before performing the linear inner product; <·,·> is the inner product operation, which is used to comprehensively measure the matching degree between the social hot spot vector and the scientific research result vector in the multidimensional space; exp(α·<·>) is the exponential kernel function, which can amplify the contribution brought by high matching degree;

[0107] α is the balancing coefficient, which is used to adjust the sensitivity of the exponential kernel function to high matching. δ(·) is the Dirac function, which ensures that it makes a non-zero contribution to the double integral only when σ = τ + Δ, thereby capturing the corresponding relationship between topic activity and scientific research results under the time lag Δ. [t0, t1] is the time observation interval, and the lower and upper limits are set according to the available data range or analysis requirements.

[0108] By the time-delayed correlation integral γ k (Δ), if the γ obtained by integrating a positive value Δ k (Δ) is significantly higher than other offsets. Other offsets refer to other possible time lag values different from the positive value Δ currently under investigation. This indicates that after this topic reaches its peak popularity on the social platform, it will show a corresponding increase or attention in formal scientific research output approximately Δ time later. This can provide a more forward-looking quantitative basis for the subsequent value assessment of emerging hot spots and scientific research management decision-making in universities.

[0109] Based on this, the potential influence score Π can be further calculated k , identify topics with shorter time lag, higher output, or topics that may not have been transformed but have huge incremental potential, and store these scoring results as time lag impact measures and potential influence scores D 402 ;

[0110] Introducing the potential influence score Π k To quantify the potential impact of hot topics on formal scientific research results (papers, project cooperation networks, etc.) under different time lag conditions, the following weighted cumulative score form can be designed:

[0111]

[0112] Where: k is the comprehensive potential influence score of hot topic k, γ k (Δ) is the double integral correlation of the topic with respect to the formal scientific research output, ψ(·) is an enhancement or suppression function used to highlight the time lag window with a very high correlation with the scientific research results (which can be set as an exponential, logarithmic or other nonlinear mapping); κ(Δ) is a time lag weighting function, which can be set to increase or decrease the weight according to actual management needs to balance the relative importance of short-term or long-term potential impacts, Δ max The maximum observation time lag can be determined based on the actual data volume and scientific research output cycle;

[0113] In this model, if the correlation between a topic’s impact on scientific research results remains high within a wide range of time lags and also increases rapidly in a short period of time, the influence score Π k Often achieve a higher score, indicating that the topic has significant potential for promoting scientific research; after completing the above calculations, the influence score of each topic can be π in the data structure k and its lag peak point Δ * Store together for easy visualization and subsequent management decision-making;

[0114] In generating the potential influence score Π k and the peak lag Δ * After that, you can further conduct multi-dimensional visualization and key topic prompts. You can refer to the following:

[0115] Timeline and results map superposition: the activity curve of each topic (k) in social media, the publication or citation curve of the corresponding scientific research results, and Δ * The significant linkage points at Δ are projected into the same coordinate system to form a time superposition diagram of influence-time lag scatter diagram: * is the horizontal axis, Π k As the vertical axis, the hot topics are scatter-visualized. The color or size can represent different academic fields or the overlap with existing literature, so as to highlight which topics have the characteristics of small lag i + high influence.

[0116] If Π k Exceeding the scoring threshold ω impact , and its Δ * Below the critical value ω delay , marked as rapid conversion type;

[0117] The pair has not yet appeared in the formal literature, but Π k Topics that are significantly higher than the average level in the same field are marked as potential explosive topics; they are highlighted or popped up in the visual interface to remind university administrators or research teams, and are stored in a structured manner (D 403or similar data objects) to write detailed indicators of such key topics, including: time lag integral curve γ k (Δ), potential influence score Π k , academic entity labels, cluster categories, etc.;

[0118] For the balance coefficient α, the maximum observation lag Δ max , scoring threshold ω impact , critical value ω delay Dynamically adjusting key hyperparameters such as [value] and [number] to observe the changes in the influence of hot topics under different time lag sensitivities allows for dynamic screening of specific subject areas or cooperative institutions based on the actual needs of institutions, facilitating managers to make more refined decisions.

[0119] When used, the time-delayed integral γ k (Δ) Quantitatively presents the potential driving effect of a topic on the output of scientific research results after its social popularity reaches its peak, and the potential influence score Π k It helps to discover cutting-edge directions that are both highly popular and rapidly transformable, and can tap into high-potential topics: it calculates correlations under different conditions, supports two-level evaluations of macro long cycles and short cycles, and helps management formulate phased scientific research strategies.

[0120] Step 403: Combine the time-delayed correlation integral γ k (Δ) and potential influence score Π k , present hot topics in a multi-dimensional way in a visual interface,

[0121] Timeline comparison chart: Displays the topic activity curve and the cumulative release curve of corresponding formal scientific research results on the same timeline, showing the lagging relationship between the two;

[0122] Topic scatter matrix: The time lag peak Δ of each topic * Projecting the impact score onto a two-dimensional or three-dimensional coordinate system allows for the identification of cutting-edge topics with short lag periods and high growth potential.

[0123] Key Tips and Highlights: For hot topics that have not yet appeared in a large number of formal documents but have continued to rise in discussion on social platforms and have high potential as shown by time-lag analysis, set prominent marks and list them as key topics. 403 , which is convenient for managers to quickly search in the system interface;

[0124] When using, comparing the topic activity curve and the official achievement release curve in the same coordinate system can help managers intuitively understand the time lag relationship. * and potential influence score Π kUsing this as a coordinate, we can quickly identify hot topics with low latency and high impact. This highlights potential explosive topics that have not yet appeared widely in formal literature but have a clear upward trend in academic research, allowing managers to intervene and provide support in advance.

[0125] This step provides a more three-dimensional explanation of early hotspots from the perspective of mapping visualization with formal scientific research results. In particular, through time-lag analysis and potential influence scoring, it links the discussion dynamics on social platforms with the actual implementation of scientific research outputs, achieving in-depth observation of academic trends and providing more forward-looking decision-making support for universities and research institutions.

[0126] Step 5: Calculate the cross-departmental synergy index based on hot topic information to identify the best teams or departments, assist in dynamic push and incentive measures, and measure potential patent coverage with the technology gap index. If the technology gap index exceeds the preset gap threshold, issue an external notification;

[0127] The step five includes the following:

[0128] Step 501: Obtain key topic list D 403 Then, we read the hot topics marked as potential explosive or rapid transformation, and use their academic field labels, time-lag impact measure γ k (Δ) and potential influence score Π k and other information to construct the cross-departmental synergy index Θ col (k,a), can be used to identify long-term or phased synergy potential, where

[0129]

[0130] Among them, S k (τ) represents the academic entity distribution vector of topic k at the normalized time τ (such as the frequency of research directions and key terms, expert contribution, etc.), M a (σ) represents the research capability vector of department a at normalized time σ (which may include multidimensional factors such as faculty structure, project resources, and scientific research equipment indicators); Φ(·) is a mapping function used to perform nonlinear transformation (such as exponential or logarithmic scale) on the department research capability vector to better match the characteristic space of the academic entity vector; <·,·> is the inner product operation used to measure the matching degree between academic entities and department capabilities in multidimensional space; exp(β·<·,·>) is the exponential kernel function, and β is the balance coefficient used to amplify the contribution brought by high matching degree; κ(τ,σ) is the time weight kernel function, which controls the coupling degree of different time segments (for example, it can decay with an increase, emphasizing strong synergy in synchronous or similar time periods); the double integral range is [0,1]×[0,1], and proportional normalization can be performed according to the data scale or time granularity;

[0131] If the cross-departmental synergy index Θ col (k,a) shows significant peaks in multiple topics, indicating that the department's competence requirements and academic elements are highly aligned with those of the specific topic over a certain period of time, demonstrating the potential for deep synergy. This double integral kernel function can quantify the degree of fit between topics and departments over a broader time dimension, facilitating subsequent phased measurement or comparison of synergy effects.

[0132] By increasing the β value, high-matching cases can be emphasized: with the help of a properly designed κ(τ,σ), the synchronization of similar time segments can be strengthened, thus providing a more forward-looking basis for cross-departmental research collaboration in large-scale, multi-time series data scenarios;

[0133] The cross-departmental synergy index Θ col (k,a) can filter the departments or teams with the highest coupling degree with topic k, and store the results uniformly as intelligent push and cross-department collaborative incentive results D 501 To achieve subsequent intelligent push, based on the cross-departmental collaboration index Θ col Based on the ranking of topics, we will recommend a list of the most promising topics to department heads or leaders of relevant disciplines, and suggest the establishment of an interdisciplinary research incentive program to promote the rapid formation of cross-domain joint teams, conduct concentrated project establishment or research on early hot topics, and achieve efficient allocation and coordination of internal university resources;

[0134] When used, targeted push of high-potential topics: by calculating the cross-departmental synergy index Θ col (k, a), automatically find the department or team that best matches the topic, which can achieve accurate push. It is recommended to set up interdisciplinary research incentives, accelerate the formation of cross-domain joint teams, improve the response speed and R&D efficiency to early hot spots, encourage interdisciplinary resource integration, combine the potential value of topics, departmental research capabilities and management decisions, form an efficient mechanism for flexible allocation of internal resources, and realize closed-loop linkage.

[0135] Step 502: Generate intelligent push and cross-department collaborative incentive results D 501 After that, for the intelligent push and cross-departmental collaborative incentive results D 501 For the hot topic k, search for existing patents or patent application documents in the same field and calculate the technology gap index Ω gap (k) Used to judge patent coverage density and innovation space, as follows:

[0136]

[0137] Where: E k A multidimensional concept vector representing topic k (which can be generated by models such as Transformer), which integrates key terms, research paradigms, academic keywords, and other features. Extract the same text or technical features for the concept vector of the i-th patent document, is a distance metric function, which can be Wasserstein distance, Jensen-Shannon divergence (JSD), KL divergence, etc. It is an exponential mapping of distance, which is used to amplify the influence when the concept distance is small, and weaken it when it is small. All patent documents are accumulated to reflect the overall density of patent coverage; μ and χ are balance coefficients that control the overall attenuation rate. If the average distance to the existing patent vector group is generally small, the gap index value will decrease accordingly, indicating that the patent coverage is dense. N is the size of the patent or patent application documents in the field.

[0138] If the distance between topic k and the existing patent document set is generally small ( The value is low), the cumulative value increases rapidly, and then through the outer layer exp (-β...) makes the technical gap index Ω gap (k) decreases significantly, indicating that the innovation gap is not large. If it is difficult to find similar concept vectors in the current patent documents for this topic ( The value is generally high), the cumulative result is small, and the technical gap index Ω gap (k) A relatively higher value indicates that there is more room for technological breakthroughs due to insufficient patent coverage;

[0139] If the technical gap index Ω gap (k) If the number of patents is higher than the preset gap threshold, it indicates that there is still room for development in the existing patent landscape. It is possible to consider applying for core patents or establishing a key technology protection network to send a warning to the outside world;

[0140] Construct corresponding demand vector D for external entities such as cooperative enterprises and industry associations v , combined with S k (t) Perform inner product matching on the academic entity distribution to identify highly compatible industrial partners; push shareable early research results to external units, invite them to jointly establish industrial application joint laboratories or incubation platforms, conduct rapid technical trials and transfer the results to the market, and ultimately form information on the progress of industry-university-research incubation. 502 , record the layout progress of each hot topic at the level of patents and industry-university-research docking and the progress of external cooperation, and provide a continuous tracking and feedback mechanism for university management and external investment decision-making;

[0141] When used, the technical gap index Ω gap(k) This system measures topics and existing patent coverage, identifying areas with potential for innovation and facilitating early deployment of core patents. It also connects industry, academia, and research: This system integrates industry partners' needs with academic entities, quickly identifying collaborative opportunities, and incubating technological achievements. This system provides comprehensive support for practical application scenarios. It not only enables rapid response to early academic signals but also extends research results into patents and industry incubation, significantly enhancing universities' resource allocation capabilities and overall competitiveness in cutting-edge technology.

[0142] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0143] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0144] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only for some logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0145] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0146] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An economic return prediction method driven by academic hot spots, characterized by: include, We crawled original posts from multi-source academic social platforms, identified duplicate content using a multi-dimensional similarity function, merged or deleted content if the MDI value exceeded the corresponding threshold, performed anomaly detection and noise filtering on the remaining text, and output a structured preliminary cleansed dataset. Based on the pre-trained BERT model and customized academic dictionary, we perform academic entity recognition and multi-dimensional text judgment on each cleaned data item, remove irrelevant information, and perform fused semantic vector mapping on the retained academic text to obtain a collection of academic vectorized texts. At the global level of the academic vectorized text collection, we extract major categories, subdivide these categories into subcategories, and generate preliminary cluster labels. We then perform activity statistics on topics over a time window and detect sudden increases using a cumulative increment function. If the activity increment threshold and academic value threshold are met, we output the corresponding early hotspot. We map early hot topics with existing formal scientific research results in a two-way manner, and use the time-lag correlation integral to measure the linkage between topics and scientific research outputs after the peak of popularity, combined with the potential influence score Π k , measure its translation potential and finally present the time-lagged evolution and high-impact topics in a visual way; Based on hot topic information, the cross-departmental synergy index is calculated to identify the best teams or departments, assist in dynamic push and incentive measures, and measure potential patent coverage with the technology gap index. If the technology gap index is higher than the preset gap threshold, an external prompt will be issued.

2. The method for predicting economic returns driven by academic hot spots as claimed in claim 1, characterized in that: According to the predefined target academic topics or keywords, corresponding posts are captured from the public platform and recorded as {P i,j } and with basic meta-information, introduce the multi-dimensional similarity function MDI(P i,j ,P i,k ); If the obtained multi-dimensional similarity function MDI value is higher than the preset threshold, the redundant data is deleted and the only valid version is retained, and the fields of the remaining posts are mapped in a unified format; Combined with the noise characteristics of academic social platforms, anomaly filtering is performed, including: checking for extremely abnormal post lengths, noise patterns such as full symbols or full links, verifying text that cannot be fully parsed during the crawling process, and eliminating unreadable or severely distorted content.

3. The method for predicting economic returns driven by academic hot spots according to claim 2, characterized in that: Parse the post text one by one, using the pre-trained BERT model combined with a customized academic dictionary to perform academic entity recognition, marking features such as research direction, technical terms, and author citations as {E k }, for each post P s Generate word segmentation sequence T s , record word units and tagged academic entity location information; Perform multi-dimensional text judgment on each post, using the judgment function Ω(P s ), if Ω(P s ) returns a true value, the post is retained and output to the post collection; otherwise, it is marked as entertainment content and removed.

4. The method for predicting economic returns driven by academic hot spots according to claim 3, characterized in that: Call the same pre-trained BERT model to get the basis vector V base (P s ), integrating academic entity weights and contextual relationships to obtain the complete vector V(P s ), defined as follows: Where: V base (P s ) is the model for the post P s The global embedding of E k (P s ) represents the semantic position of the kth academic entity in the post, Φ(·) is the local context embedding function; γ k is the fusion weight.

5. The method for predicting economic returns driven by academic hot spots according to claim 4, characterized in that: For each post P s The embedding representation V(P s ) performs multi-level clustering, where a vector matrix is constructed Where n is the number of posts, m is the vector dimension, and each row corresponds to a post vector V(P s ); use pre-selected hierarchical clustering at the global level to extract several major categories, and further subdivide each major category based on the distribution of academic entities, and subdivide similar professional directions into more precise topic subcategories.

6. The method for predicting economic returns driven by academic hot spots according to claim 5, characterized in that: Perform time series statistics on each topic, record the discussion volume and interaction indicators of posts in discrete time periods, and record the comprehensive activity vector of a topic k at time τ as h k (τ), in order to detect short-term spikes, the following formula is defined as the cumulative increment function Λ of the multidimensional activity curve k (t), if the cumulative increment function Λ k (t) If a sudden jump occurs within a certain time interval, the corresponding topic is determined to be a potential early hot spot; Among them, h k (τ) represents the activity vector of topic k at time r, Its gradient at τ, w k is the weight vector corresponding to topic k, and <·,·> is the inner product operation.

7. The method for predicting economic returns driven by academic hot spots according to claim 6, characterized in that: According to the cumulative increment function Λ k (t) The maximum increment interval of the post is used to filter the post text of the corresponding sub-topic, and the sub-topics that pass the screening are marked as early signals. For topics that do not meet the threshold or have a sudden increase but low content value, they are marked as ordinary discussions; each early hot topic and its activity curve are extracted, and the key academic entities and tags of each hot topic are searched in the formal scientific research database to obtain the corresponding research results entries.

8. The method for predicting economic returns driven by academic hot spots according to claim 7, characterized in that: In order to evaluate the time lag between the appearance of hot topics on social platforms and the subsequent output of corresponding formal results, and to comprehensively score the potential influence, the following time lag-related score Υ is defined: k (Δ) is used to measure the degree of linkage between the two. The formula is as follows: Among them: k (Δ) is the time offset of topic k at Δ, z k (τ) represents the social media activity vector of topic k at time τ; r k (σ) represents the formal scientific research achievement record corresponding to topic k at time σ; Φ(·) is the high-level mapping of the formal scientific research achievement vector; <·,·> is the inner product operation, exp(α·<·>) is the exponential kernel function, α is the balance coefficient; δ(·) is the Dirac function, and [t0,t1] is the time observation interval.

9. The method for predicting economic returns driven by academic hot spots according to claim 8, characterized in that: Introducing the potential influence score Π k To quantify the potential impact of hot topics on formal scientific research results under different time lag conditions, the following weighted cumulative integral form can be designed: Where: k is the comprehensive potential influence score of hot topic k, Υ k (Δ) is the double integral relevance of the topic relative to the formal scientific research output, ψ(·) is the enhancement or suppression function; κ(Δ) is the time lag weighting function, Δ max is the maximum observation lag.

10. The method for predicting economic returns driven by academic hot spots according to claim 9, characterized in that: Generate potential influence score Π k and the peak lag Δ * Then, multi-dimensional visualization and key topic prompts are performed, in which the activity curve of each topic (k) in social media, the publication or citation curve of the corresponding scientific research results, and Δ * The significant linkage points at Δ are projected into the same coordinate system to form a time superposition diagram of influence-time lag scatter diagram: * is the horizontal axis, Π k As the vertical axis, the hot topics are scatter-visualized, and the color or size can represent different academic fields or the overlap with existing literature; If Π k Exceeding the scoring threshold ω impact , and its Δ * Below the critical value ω delay , marked as rapid conversion type; has not yet appeared in formal literature, but Π k Topics that are significantly higher than the average level in the same field are marked as potentially explosive and will be highlighted or prompted to university administrators or research teams in the visual interface through pop-up windows.

11. The method for predicting economic returns driven by academic hot spots according to claim 10, characterized in that: Combined with the time-delay dependent integral Υ k (Δ) and potential influence score Π k , hot topics are presented in a multi-dimensional way in a visual interface, where the topic activity curve and the cumulative release curve of the corresponding formal scientific research results are displayed on the same timeline, showing the lagging relationship between the two; The peak value of the time delay of each topic Δ * Together with the influence score, it is projected into a two-dimensional or three-dimensional coordinate system to focus on identifying cutting-edge topics with short lag periods and high growth potential; hot topics that have not yet appeared in a large number of formal literature but have continued to increase in discussion on social platforms and whose time-lag analysis shows high potential are marked significantly.

12. The method for predicting economic returns driven by academic hot spots according to claim 11, characterized in that: Read hot topics marked as potentially explosive or rapidly transforming, based on their academic field labels and time-lag impact measures k (Δ) and potential influence score Π k Information, construct cross-departmental synergy index Θ col (k,a), through the cross-departmental synergy index Θ col (k,a) selects the departments or teams with the highest coupling degree with topic k, and uses the cross-departmental synergy index Θ col The ranking of the most promising topics will be pushed to the heads of departments or relevant disciplines; among them, Among them, S k (τ) represents the academic entity distribution vector of topic k at normalized time τ, M a (σ) represents the research capability vector of department a at normalized time σ; Φ(·) is the mapping function, <·,·> is the inner product operation, exp(β·<·,·>) is the exponential kernel function, β is the balance coefficient; κ(τ,σ) is the time weight kernel function.

13. The method for predicting economic returns driven by academic hot spots according to claim 12, characterized in that: For hot topic k, search existing patents or patent application documents in the same field and calculate the technology gap index Ω gap (k) Used to judge patent coverage density and innovation space, as follows: Where: E k Represents the multidimensional concept vector of topic k, is the concept vector of the i-th patent document, is the distance measurement function; construct the corresponding demand vector D for the external subject v , combined with S k (t) performs inner product matching on the academic entity distribution to identify highly compatible industrial partners.

Images