Scientific research performance information analysis and evaluation system and processing method thereof

Abstract

The invention belongs to the technical field of computer information processing, and relates to a scientific research performance information analysis and evaluation system and a processing method thereof.The scientific research performance information analysis and evaluation system comprises a data collection module, a weight distribution module, a real-time feedback module and an execution engine function module; weight distribution receives the first signal and analyzes a current strategic target priority identifier and a historical performance evaluation record contained in the first signal, a weight parameter space dynamically matched with a target priority is constructed through an online learning algorithm, and a second signal containing a real-time weight mapping relation is generated; real-time feedback is carried out, a multi-dimensional performance evaluation matrix is constructed based on the weight mapping relation in the second signal, and a third signal is generated; the execution engine receives the third signal, analyzes an abnormal positioning code of the third signal, and aggregates intervention records in a period to form a fourth signal; according to the method, the defects that traditional performance information analysis depends on static weight, the data dimension is single, and real-time dynamic adjustment is lacked can be overcome, and the analysis efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the field of computer information processing technology, and relates to a scientific research performance information analysis and evaluation system and a processing method thereof that adopts modern computer information processing functions. The system can dynamically evaluate and manage scientific research performance through multi-source data combined with real-time adjustment of evaluation results, and simultaneously generate personalized intervention strategies. Background Art

[0002] Generally, performance evaluation refers to a comprehensive assessment of the degree to which central government departments have achieved their performance targets, as well as the execution of the budget allocated to achieve these targets, using specific evaluation methods, quantitative indicators, and standards. The performance evaluation process compares an employee's actual salary performance with the required performance standards. This helps understand the training and education needs of employees and teams, and provides information for work plans, budget assessments, and human resource planning. Traditional performance evaluation systems generally use fixed-weight models, which are unable to adapt to the dynamic changes in corporate strategic objectives. Existing technologies primarily rely on structured data and lack the ability to process diverse information such as unstructured text and sensor behavior, resulting in a single evaluation dimension and low data utilization. Excessive weighting of historical data can lead to delayed evaluation results, making it difficult to timely reflect employees' real-time work status. Furthermore, the system lacks an effective feedback loop mechanism, and the generation of intervention measures relies on manual judgment, resulting in slow response and insufficient personalization. Although some technologies have attempted to incorporate machine learning algorithms, the models lack interpretability and are loosely integrated with specific business scenarios, making it difficult to dynamically optimize weightings. At the data integration level, the format conversion and correlation analysis of heterogeneous data across systems are inefficient, making it impossible to build a cross-modal data correlation graph. Regarding privacy protection, most systems rely solely on permission management and fail to implement feature desensitization during the data collection phase.

[0003] In the prior art, Chinese patent CN110033191B discloses a business artificial intelligence analysis method and system, comprising the following steps: S1. Establishing a performance evaluation and improvement system within the business artificial intelligence system, which includes BI analysis and performance evaluation and improvement functions; S2. Utilizing the performance evaluation and improvement system, a knowledge base is built through machine learning, and recommendations for management improvement measures are provided based on actual performance. Advantages include: relying on a performance evaluation and improvement model to record, evaluate, and improve the performance of organizational members; aggregating and decomposing the performance of organizational members enables comprehensive recording, evaluation, and improvement of the overall organizational performance; establishing multiple, interconnected performance evaluation and improvement processes to drive continuous improvement and achieve organizational goals; and establishing a machine learning process of "organization-performance status-solution-effect" to enable self-learning and intelligent recommendation of problem-solving measures. However, this patent relies on a preset performance evaluation model, resulting in fixed weights that cannot be dynamically adjusted. While its machine learning knowledge base can recommend improvement measures, it fails to address the real-time matching of weights with strategic objectives, causing the recommended measures to lag behind changes in scientific research objectives.

[0004] Chinese patent CN118822342A discloses a system, method, and terminal for processing employee data points. The system includes an information modeling module, an operation module, a statistics module, and an alarm module. The information modeling module constructs an employee work information model by collecting past employee work information. The operation module is used to implement the interaction between performance and points based on work items. An inspection period is set, denoted as t. The statistics module is used to calculate the points of individual employees within the inspection period to obtain a numerical value. The alarm module compares the past points with the obtained points and generates an execution instruction based on the comparison result. The information modeling module includes an HR information recording unit, a collation unit, and an information interaction unit. Through the cooperation of these modules, a clear reward mechanism can be provided to motivate employees to achieve performance goals, while the enterprise can more objectively and fairly evaluate employee work performance. However, its scoring rules rely on preset thresholds, and the weighting system is static. The alarm module can only compare historical points and cannot dynamically adjust the score calculation rules according to the urgency of scientific research projects. This leads to a disconnect between incentive measures and real-time needs, and it is impossible to adjust and improve the real-time allocation of measures.

[0005] Chinese patent CN108038583A discloses a method and system for dynamically managing an enterprise's performance indicators to address facility operations management goals. The method includes integrating operational data associated with the enterprise; deriving key performance indicators (KPIs), thresholds, and metrics; determining factors that affect the performance of the KPIs and factors that affect the performance of the thresholds; normalizing the factors that affect the performance of the KPIs based on a comparability matrix; evaluating the performance of the KPIs based on the one or more thresholds, the comparability matrix, and the associated patterns of the normalized data to derive a KPI performance analysis; fine-tuning the KPIs based on one or more usage patterns of interactive visualization and fine-tuning and enhancing the system, the one or more KPIs, and the interactive visualization, thereby dynamically managing the enterprise's performance indicators to address facility operations management goals. However, its threshold setting relies on manual pre-setting, and the automatic calibration of factor weights is not achieved, resulting in KPI evaluation being difficult to adapt to the nonlinear characteristics of scientific research scenarios.

[0006] Chinese patent CN114897363A discloses a business management method based on big data analysis, relating to the field of business management technology, and comprising the following steps: S1: registering and recording the information of business employees and recording the information in an analysis module; S2: recording the attendance of business employees through an attendance module, which can send the attendance data to an analysis host. This invention uses an analysis module to analyze the working hours of a month, then obtains the average work efficiency of employees of similar businesses through big data, and then calculates the business's task volume for that month. During subsequent work, the recording module records the daily workload of each employee, and at the end of the month, the analysis module calculates the work efficiency of each employee. When the business's task volume for that month is subsequently calculated again, the calculation can be based on the employee's work efficiency, so that the calculation result is more accurate, preventing fluctuations in business management and facilitating management by managers. However, it only integrates attendance and basic task volume data, and its analysis module relies on external big data, lacking in-depth exploration of internal scientific research innovation or resource utilization.

[0007] Chinese patent CN118036965A discloses an intelligent enterprise information management system, particularly in the field of smart enterprise technology. The system includes an information acquisition module for acquiring employee attendance and work information; an attendance analysis module for analyzing employee clock-in and attendance anomalies; an efficiency analysis module for analyzing employee idle time and efficiency indexes and adjusting the analysis process; a management module for assigning work tasks to employees and evaluating their work performance; a management optimization module for optimizing the employee work performance evaluation process; and an update module for updating the employee efficiency index for the next management cycle based on the average of the actual completion time of each work task within the management cycle. This invention improves the efficiency of employee work management and enterprise information management. However, the management optimization module requires waiting for the end of the management cycle to update the efficiency index, making the cycle too rigid. Its evaluation optimization relies on historical averages, making it impossible to intervene in anomalies in real time during the research process, and the evaluation results too rigid.

[0008] After searching and analyzing, the inventors found that no existing technology has yet disclosed a scientific research performance evaluation system and its scientific processing and management methods that can deeply integrate multi-source data, adjust evaluation weights in real time, automatically generate personalized intervention strategies, and form an optimized closed-loop intelligent system. Therefore, the innovation sought involves a scientific research performance information analysis and evaluation system and its processing methods, which can improve the shortcomings of existing technologies and solve the problems of traditional scientific research performance information analysis and evaluation systems that rely on static weights, have a single data dimension, and lack real-time dynamic adjustment capabilities, thereby realizing real-time intelligent performance analysis. Summary of the Invention

[0009] The purpose of the present invention is to overcome the shortcomings of the existing technology, seek innovative design and provide a scientific research performance information analysis and evaluation system and its processing method, which are used to solve the problems of traditional scientific research performance information analysis and evaluation systems relying on static weights, single data dimensions and lack of real-time dynamic adjustment capabilities; the problem is solved by constructing a dynamic weight distribution module and a real-time feedback intervention mechanism: the data acquisition module integrates multi-source heterogeneous data such as attendance records, task status, text interaction and sensor behavior, and generates a first signal with a timestamp after standardized processing; the weight distribution module uses an online learning algorithm to parse the strategic goal priority, dynamically constructs a weight parameter space, reduces the impact of historical data through a preset attenuation factor, and generates a second signal for real-time weight mapping; the real-time feedback module constructs a multi-dimensional evaluation matrix based on the weight relationship, monitors indicator deviations and triggers a third signal containing abnormal codes and improvement measures; the execution engine module generates personalized plans through similarity matching, aggregates intervention records to form a fourth signal to feed back to the weight distribution module, forming a closed-loop system of dynamic weight adjustment and execution optimization, thereby realizing strategic-oriented precise evaluation and automated intervention.

[0010] In order to achieve the above-mentioned purpose of the invention, the present invention provides a scientific research performance information analysis and evaluation system, whose functional structure includes: a data acquisition module forms a first signal containing a timestamp mark; a weight allocation module receives the first signal and parses the current strategic goal priority identifier and historical performance evaluation records contained therein, and constructs a weight parameter space that dynamically matches the strategic goal priority through an online learning algorithm, wherein each weight coefficient in the weight parameter space automatically reduces the influence of historical data over time according to a preset attenuation factor, and generates a second signal containing a real-time weight mapping relationship; the real-time feedback module constructs a multidimensional performance evaluation matrix based on the weight mapping relationship in the second signal, and generates a third signal containing an anomaly location code, a recommended measure list and a trigger time node; the execution engine module receives the third signal and parses the anomaly location code contained therein, performs similarity matching on the recommended measure list with the employee capability archive to generate a personalized improvement plan, and at the same time aggregates all intervention records in the current cycle to form a fourth signal, and the fourth signal is fed back to the weight allocation module to drive the iterative optimization of the weight parameter space.

[0011] The data acquisition module involved in the present invention includes a multi-source data fusion engine. The multi-source data fusion engine converts heterogeneous data formats for structured attendance data from an enterprise resource planning system, semi-structured task logs from an office automation system, unstructured text interaction content collected by employee terminal devices, and time-series behavior data generated by wearable device sensors, establishes a cross-modal data association relationship map, and maps key indicators representing work engagement in different data sources to a unified vector space through a feature extractor. After completing data alignment, the multi-source data fusion engine injects a timestamp mark to form a first signal, wherein the unstructured text interaction content is divided into dialogue topic paragraphs and attached with participant identifiers through a semantic segmentation unit, and the sensor behavior data generates a periodic behavior pattern feature vector through a sliding window statistical method.

[0012] The data acquisition module involved in the present invention further includes a natural language processing unit, which performs dependency syntax analysis and sentiment polarity calculation on the text interaction content, extracts the keyword frequency distribution matrix related to task allocation in the dialogue segment, identifies the innovative suggestions put forward by the speaker and constructs a contribution intensity index. The natural language processing unit captures the contextual association relationship in the long text through the attention mechanism, generates an additional feature vector containing a collaboration score and injects it into the first signal set, wherein the collaboration score is weightedly calculated based on the number of times the topic is actively raised in the dialogue round and the frequency of the solution being adopted, and the sentiment polarity calculation result is used to correct the semantic weight of the text interaction content.

[0013] The weight distribution module involved in the present invention also includes a dynamic adjustment unit. The attenuation factor preset in the weight distribution module realizes nonlinear changes through the dynamic adjustment unit. The dynamic adjustment unit receives the target change notification signal from the enterprise strategic management system. When it is detected that the strategic priority has switched across dimensions, the attenuation factor recalibration process is started. The transition curve equation of the new and old weights is calculated based on the correlation matrix before and after the strategic goal switching. The attenuation factor recalibration process includes the gradual attenuation of the historical weight influence and the exponential smoothing introduction of the new strategic weight. After completing the parameter update, the dynamic adjustment unit generates a weight migration trajectory report and writes it into the system audit log.

[0014] The online learning algorithm in the weight allocation module involved in the present invention adopts a two-layer optimization framework, wherein the first-layer optimizer is configured to construct an initial weight parameter space according to the real-time strategic goal priority, and the second-layer optimizer is configured to receive the intervention effect feedback data carried in the fourth signal from the execution engine module, and adjust the loss function constraint conditions of the first-layer optimizer through the backpropagation mechanism. The two-layer optimization framework realizes the coordinated optimization of strategic guidance and execution feedback through alternating iterations, wherein the second-layer optimizer calculates the elasticity coefficient of the intervention measures on the performance evaluation index during each iteration, and injects the coefficient as a regularization term into the objective function of the first-layer optimizer.

[0015] The multidimensional performance evaluation matrix in the real-time feedback module involved in the present invention includes a longitudinal time dimension and a transverse capability dimension. The longitudinal time dimension divides the evaluation cycle through a sliding time window and establishes a historical performance trend regression model. The transverse capability dimension constructs a capability item association rule base based on the job description. The deviation degree detection unit is configured to calculate the standardized residual value of each capability item of the individual indicator in the transverse dimension. When the residual value of the same capability item exceeds two standard deviations of the team benchmark value in three consecutive evaluation cycles, a progressive intervention strategy is activated. The progressive intervention strategy includes three levels of progressive measures: early warning notification, targeted training suggestions, and job adaptability adjustment plan.

[0016] The real-time feedback module involved in the present invention further includes a situational awareness unit, which is configured to receive industry volatility index and market competition intensity data from the enterprise's external environment monitoring system, build a dynamic threshold adjustment model, and automatically lower the deviation detection threshold sensitivity when it is detected that the external environment change rate exceeds a preset critical value. The situational awareness unit extracts the periodic fluctuation characteristics in the external environment data through a convolutional neural network, and encodes it as an environmental pressure coefficient and injects it into a weight correction item of the multidimensional performance evaluation matrix, wherein the elastic relationship between the environmental pressure coefficient and the team baseline value is modeled through a multivariate regression equation.

[0017] The similarity matching process in the execution engine module involved in the present invention adopts a multimodal embedding model. The model is configured to map the text descriptions of skill certificates, project experience structured data and historical training records in the employee capability archive to a unified semantic space to generate employee capability feature vectors. At the same time, the intervention measure text in the recommended measure list is encoded into sentence vectors, the cosine similarity between the two is calculated, and the job competency model is introduced as a constraint condition. When generating personalized improvement plans, the multimodal embedding model retains the top K optimal matching results and adds a feasibility assessment score. The feasibility assessment score is dynamically adjusted according to the ratio of the implementation cost matrix to the expected benefit matrix.

[0018] The execution engine module involved in the present invention further includes an intervention effect tracking unit, which is configured to start the effect monitoring process after sending the personalized improvement plan, calculate the effectiveness coefficient of the measure by comparing the change rate of the performance evaluation index of the target capability item before and after the intervention, and automatically trigger the alternative plan generation mechanism when it is detected that the effectiveness coefficient is lower than the preset level. The intervention effect tracking unit inputs the effectiveness coefficient sequence in each cycle into the time series prediction model, generates the effect attenuation curve of the next three evaluation cycles, and optimizes the parameter settings of the similarity matching algorithm accordingly.

[0019] The present invention further designs and provides a method for analyzing and evaluating scientific research performance information by means of a modern computer information processing system. The specific processing steps include: S1: Information acquisition and vectorization processing: Acquire multi-source heterogeneous data including attendance records, text interaction content, and sensor behavior data, and perform standardization cleaning and feature vectorization processing on the multi-source heterogeneous data to form a first signal with a timestamp mark; S2: Constructing a weight parameter space: Receive the first signal and parse the current strategic goal priority identifier and historical performance evaluation records contained therein. Use an online learning algorithm to construct a weight parameter space that dynamically matches the strategic goal priority. Each weight coefficient in the weight parameter space automatically reduces the influence of historical data over time according to a preset attenuation factor, generating a second signal containing a real-time weight mapping relationship. S3: Constructing a performance evaluation matrix: Constructing a multi-dimensional performance evaluation matrix based on the weight mapping relationship in the second signal, and generating a third signal containing an anomaly location code, a list of recommended measures, and a trigger time node; S4: Optimize weight parameter space: Receive the third signal and parse the abnormal location code contained in it, perform similarity matching between the list of recommended measures and the employee capability archive to generate a personalized improvement plan, and at the same time aggregate all intervention records in the current cycle to form the fourth signal to drive the iterative optimization of the weight parameter space.

[0020] Compared with the existing technology, the present invention provides a scientific research performance information analysis and evaluation system and a processing method thereof, which solves the problem by constructing a dynamic weight allocation module and a real-time feedback intervention mechanism: the data acquisition module integrates multi-source heterogeneous data such as attendance records, task status, text interaction and sensor behavior, and generates a first signal with a timestamp after standardized processing; the weight allocation module uses an online learning algorithm to analyze the strategic goal priority, dynamically constructs a weight parameter space, reduces the influence of historical data by a preset attenuation factor, and generates a second signal of real-time weight mapping; the real-time feedback module constructs a multi-dimensional evaluation matrix based on the weight relationship, monitors indicator deviations and triggers a third signal containing abnormal coding and improvement measures; the execution engine module generates personalized plans through similarity matching, aggregates intervention records to form a fourth signal to feed back to the weight allocation module, forming a closed-loop system of dynamic weight adjustment and execution optimization, thereby realizing strategic-oriented accurate evaluation and automated intervention; its system structure layout is scientific and reasonable, physical space is saved, and the functional application environment is friendly. It has fast processing speed and high efficiency for scientific research performance information, and the conclusions are accurate and highly available. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a schematic diagram of the system framework of the scientific research performance information analysis and evaluation system involved in the present invention.

[0022] Figure 2 This is a schematic diagram of the process flow for analyzing, evaluating and processing scientific research performance information involved in the present invention. DETAILED DESCRIPTION

[0023] The present invention is described in detail below through embodiments with reference to the accompanying drawings.

[0024] Example 1: The diagrams provided in this embodiment are only used to schematically illustrate the basic concept of the present invention. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail. Figure 1-2 shown.

[0025] This embodiment involves a scientific research performance information analysis and evaluation system, whose main functional structure includes a data acquisition module, a weight allocation module, a real-time feedback module and an execution engine module. The data acquisition module forms a first signal containing a timestamp mark; the weight allocation module receives the first signal and parses the current strategic goal priority identifier and historical performance evaluation records contained therein, and constructs a weight parameter space that dynamically matches the strategic goal priority through an online learning algorithm, wherein each weight coefficient in the weight parameter space automatically reduces the influence of historical data over time according to a preset attenuation factor, and generates a second signal containing a real-time weight mapping relationship; the real-time feedback module constructs a multidimensional performance evaluation matrix based on the weight mapping relationship in the second signal, and generates a third signal containing an anomaly location code, a recommended measure list and a trigger time node; the execution engine module receives the third signal and parses the anomaly location code contained therein, performs similarity matching on the recommended measure list with the employee capability archive to generate a personalized improvement plan, and at the same time aggregates all intervention records in the current cycle to form a fourth signal, and the fourth signal is fed back to the weight allocation module to drive iterative optimization of the weight parameter space.

[0026] This embodiment relates to a scientific research performance information analysis and evaluation system, which is composed of a functional electronic or connected combination of a data acquisition module, a weight distribution module, a real-time feedback module, and an execution engine module. The data acquisition module is responsible for acquiring multi-source heterogeneous data from enterprise information systems, employee terminal devices, and external database interfaces, including but not limited to attendance records, task completion status, text interaction content, and behavioral data generated by sensors. The enterprise information system includes an enterprise strategic management system, an enterprise resource planning system, and an office automation system. These raw data undergo a standardized cleaning process to remove noise data and redundant information, and are then converted into a numerical feature set with a unified dimension through feature vector quantization. A timestamp accurate to the millisecond level is injected into each feature vector, ultimately forming a first signal set. The signal set is transmitted to the weight distribution module via an internal communication protocol. After receiving the first signal, the weight distribution module first parses the current strategic goal priority identifier embedded therein. The identifier is usually dynamically updated by the enterprise strategic management system. For example, when the enterprise's quarterly strategic option switches from "market expansion" to "cost optimization," the strategic goal priority identifier will carry a new priority code for the new enterprise quarterly strategic option. At the same time, the weight allocation module calls the historical performance evaluation record database to extract the performance indicator completion data within the past six evaluation cycles; based on the above input, the weight allocation module uses an online learning algorithm to construct a dynamic weight parameter space. The weight parameter space contains weight coefficients, where each weight coefficient corresponds to the evaluation weight of a specific performance indicator, such as sales achievement rate, customer satisfaction, innovation contribution, etc.; the online learning algorithm can analyze the changing trend of strategic goal priorities in real time, and dynamically adjust the distribution of weight coefficients through the gradient descent method to ensure that the weight parameter space is always synchronized with the strategic orientation; each weight coefficient in the weight parameter space is configured with a preset attenuation factor, which decays exponentially over time, so that the influence of historical data gradually decreases. For example, the weight of sales data three months ago will automatically drop to 30% of the current value, thereby avoiding excessive interference of historical data on real-time evaluation.

[0027] After completing the weight calculation, this embodiment generates a second signal containing a real-time weight mapping relationship, which is transmitted to the real-time feedback module through an encrypted channel. After receiving the second signal, the real-time feedback module first parses the weight mapping relationship and constructs a multi-dimensional performance evaluation matrix. The dimension design of the matrix includes a vertical time axis and a horizontal capability item axis. The vertical dimension divides the evaluation cycle through a sliding time window mechanism, for example, the evaluation data is rolled up in a window of two weeks. The horizontal dimension establishes an evaluation benchmark based on the capability items defined in the job description (such as technical capability, collaboration capability, and innovation capability). Each cell in the matrix stores the evaluation value of the corresponding capability item within a specific time window. The deviation detection unit built into the real-time feedback module continuously monitors the difference between individual assessment values and team benchmark values. When it is detected that the deviation value of a specific capability item exceeds a preset threshold (for example, twice the standard deviation of the team average), a three-level intervention mechanism is triggered; the intervention mechanism responds according to the severity of the deviation: a level one deviation triggers a level one intervention mechanism to issue an internal system warning notification; a level two deviation triggers a level two intervention mechanism to start the targeted training recommendation generation process; a level three deviation triggers a level three intervention mechanism to generate a job suitability adjustment plan, and the degree of deviation is determined according to the preset threshold; finally, the module encapsulates the abnormal location code, the recommended measures list and the trigger time node into a third signal, and transmits it to the execution engine module through the message queue; after receiving the third signal, the execution engine module first parses the abnormal location code to determine the specific capability item and related employees that require intervention. Then, the intervention strategy template in the preset rule library is called. For example, for the exception of "insufficient customer service response speed", the rule library may contain strategy templates such as "speech optimization training", "workflow reconstruction", and "performance coaching plan"; the module matches the recommended measures list with the employee capability archive for similarity. The employee capability archive stores each employee's skill certificates, project experience, training records and other structured and unstructured data; a multimodal embedding model is used in the matching process to map the skill information described in the text and the numerical project data into a unified semantic space, calculate the cosine similarity between the recommended measures and the employee's capability characteristics, and screen out the top ones with the highest matching degree. Three plans are generated with an additional feasibility assessment score. For example, for employees with a data analysis background but lacking communication skills, the system may recommend "data visualization reporting training" rather than general communication courses. After generating a personalized improvement plan, the execution engine module aggregates all intervention records in the current evaluation cycle, including the type of measure, execution status and preliminary effect data, to form a fourth signal that is fed back to the weight allocation module. This feedback signal drives the iterative optimization of the weight parameter space. For example, when a certain type of intervention measure has a significant effect in three consecutive cycles, the system will automatically increase the weight coefficient of the relevant capability item, thereby strengthening the influence of this dimension in subsequent evaluations.

[0028] The specific implementation of the limited data collection module involved in this embodiment focuses on the design of a multi-source data fusion engine. This engine can process structured attendance data from enterprise resource planning systems, such as daily clock-in records and leave approval process data; semi-structured task logs from office automation systems, including task creation time, responsible person, completion status, and approval opinions; unstructured text interaction content collected by employee terminal devices, such as chat logs, email bodies, and meeting minutes from internal enterprise communication software; and time-series behavioral data generated by wearable device sensors, such as location movement trajectories, heart rate variability data, and keyboard stroke frequency collected by smart work badges. The multi-source data fusion engine first converts heterogeneous data formats and aligns the time series of different data sources. For example, the occurrence time of text interaction content is matched with the timeline of sensor behavior data. For unstructured text data, the engine's built-in semantic segmentation unit divides long text into independent paragraphs based on conversation topics. For example, the text of a one-hour meeting recording can be segmented into topic blocks such as "Project Progress Discussion," "Technical Difficulty Analysis," and "Resource Coordination Request." Participant identifiers are attached to each paragraph to facilitate subsequent contribution analysis. Sensor behavior data is processed using a sliding window statistical method. For example, using a 15-minute window to count employee distance traveled, duration of sitting, and number of equipment operations, this generates a feature vector representing periodic behavior patterns. The feature extractor maps this processed data into a unified vector space. For example, it converts "number of tardiness" in attendance data into a normalized value between 0 and 1, and converts "frequency of occurrence of task assignment keywords" in textual interactions into a weighted value calculated by term frequency minus inverse document frequency. After feature fusion, the engine injects millisecond-accurate timestamps into the data stream to form the first signal set. For example, for a manufacturing company, the data collection module aligns and fuses the equipment operation frequency of its production line sensors (collected every 5 seconds), the work order completion status of its MES system (updated every minute), and the discussion logs of production issues on its WeChat account (collected in real time). This generates a composite feature vector including timestamps. The feature dimensions cover evaluation metrics such as "operational standardization," "task execution efficiency," and "frequency of cross-departmental collaboration."

[0029] The data collection module involved in this embodiment incorporates a natural language processing unit to enhance in-depth analysis of textual interactions. This unit first performs dependency syntactic analysis on the text content, identifying the subject-verb-object structure and modifier relationships within the sentence. For example, in the sentence "Engineer Zhang proposed using a deep learning algorithm to optimize the quality inspection process," the unit extracts "propose" as the core action, "Engineer Zhang" as the subject, "deep learning algorithm" as the method, and "optimize the quality inspection process" as the purpose. Simultaneously, the unit performs sentiment polarity calculation, using a pre-trained model to determine the sentiment of the text segment. For example, it labels "Customers repeatedly complain about delivery delays" as negative and "The team exceeded quarterly targets" as positive. At the task allocation analysis level, the unit constructs a keyword frequency distribution matrix, for example, counting the frequency of task-related terms such as "urgent," "priority," and "deadline" to identify high-load task periods. To identify innovative suggestions, the unit combines pattern matching with semantic similarity. For example, it detects sentences containing trigger words such as "suggestion," "trial," and "optimization plan," compares their similarity with a database of historical innovation proposals, and constructs a contribution intensity index, which is weighted based on the number of times a suggestion has been adopted and the effectiveness of its implementation. To assess collaboration, the unit uses an attention mechanism to analyze contextual relationships within long texts. For example, in cross-departmental meeting minutes, it identifies instances where an employee proactively raised issues and offered solutions multiple times. The unit then calculates the percentage of their speaking turns relative to the total number of discussions, as well as the frequency with which their proposed solutions were marked "adopted," ultimately generating a collaboration score. The sentiment polarity calculation results are used to adjust the semantic weight of the text content. For example, feedback expressing negative sentiment will be given a higher weight in the assessment, prompting the system to prioritize relevant intervention processes. After processing, the natural language processing unit injects additional feature vectors, such as the collaboration score and contribution strength index, into the first signal set. For example, in data from an internet company, the data collection module analyzed the project team's Slack chat logs and identified that a team member proactively proposed technical solutions eight out of ten discussions, five of which were incorporated into the development plan. Based on this, the system generated an "Innovation Contribution Index" of 0.86 (out of a maximum score of 1.0) for this team member and fed this into the weight assignment module as a key feature.

[0030] This embodiment relates to a method for analyzing and evaluating scientific research performance information, and its specific operation and running include the following process steps: S1: Acquire data and generate a first signal: Acquire multi-source heterogeneous data including attendance records, text interaction content, and sensor behavior data, and perform standardization cleaning and feature vectorization processing on the multi-source heterogeneous data to form a first signal containing a timestamp; S2: Constructing a weight parameter space and generating a second signal: Receive the first signal and parse the current strategic goal priority identifier and historical performance evaluation records contained therein, and construct a weight parameter space that dynamically matches the strategic goal priority through an online learning algorithm, wherein each weight coefficient in the weight parameter space automatically reduces the influence of historical data over time according to a preset attenuation factor, thereby generating a second signal containing a real-time weight mapping relationship; S3: Constructing an evaluation matrix and generating a second signal: constructing a multi-dimensional performance evaluation matrix based on the weight mapping relationship in the second signal, and generating a third signal including an abnormality location code, a list of recommended measures, and a triggering time node; S4: Generate an improvement plan and generate a fourth signal: Receive the third signal and parse the abnormal location code contained therein, perform similarity matching on the list of recommended measures with the employee capability archive to generate a personalized improvement plan, and at the same time aggregate all intervention records in the current cycle to form a fourth signal to drive the iterative optimization of the weight parameter space.

[0031] This embodiment addresses the problems of indicator solidification, feedback lag, and strategy rigidity in the traditional scientific research performance evaluation system, and constructs an intelligent evaluation system with dynamic adaptability. The system architecture adopts a four-layer progressive design, including a data acquisition layer, a strategy matching layer, an analysis and decision-making layer, and an optimization feedback layer, and realizes the coordinated evolution of the evaluation model and strategic goals through a closed-loop control mechanism. In terms of technical principles, it integrates time series analysis, natural language processing, pattern recognition, and online learning algorithms. Through heterogeneous data fusion and dynamic parameter adjustment, it solves the problem of multi-dimensional indicator weight distribution and realizes the automation of the entire process from data collection to strategy optimization. Step S1 connects to the enterprise human resources system, office collaboration platform, and Internet of Things terminal devices through a distributed network interface to obtain three core data types: attendance record data includes employee on-the-job status, working hours, and abnormal sign-in records; text interaction data covers email exchanges, instant messaging records, and document collaboration content; sensor behavior data includes workstation stay time, equipment usage frequency, and movement trajectory information. The data standardization process utilizes a three-stage mechanism. After the raw data is formatted and standardized, it enters the outlier detection phase, where a sliding window algorithm is used to identify and remove noise data. The feature vectorization process employs differentiated processing strategies for different data types. Time series data uses discrete wavelet transforms for time-frequency domain feature extraction, text data uses semantic role annotation to construct a topic vector space, and behavioral data uses Markov chains to model state transition probabilities. Step S2 employs an incremental update strategy, using an online learning algorithm to decompose strategic objectives into a set of quantifiable feature dimensions. Each weight coefficient corresponds to a contribution assessment of a specific strategic indicator. The system monitors strategic priority adjustment signals in real time. When a strategic objective change is detected, the partial derivatives of the parameters for each dimension are recalculated using a gradient descent algorithm to achieve a smooth transition in the weight space. The historical data influence attenuation mechanism is designed as a dual-channel adjustment model: an explicit attenuation factor acts on the time dimension, exponentially reducing the computational weight of historical data; an implicit attenuation factor automatically weakens historical features with low correlation to the current strategic objective through feature correlation analysis. The weight mapping visualization module utilizes high-dimensional data dimensionality reduction techniques to transform the abstract parameter space into an interpretable two-dimensional weight distribution graph.

[0032] Step S3 in this embodiment consists of three orthogonal dimensions: the time dimension divides evaluation units according to work cycles, the capability dimension corresponds to the job competency model, and the behavior dimension reflects actual work performance. Matrix operations utilize tensor decomposition techniques to map raw data into low-rank subspaces for similarity calculation. The anomaly localization algorithm, based on the isolation forest principle, constructs random hyperplanes to partition the data space and identify anomalous individuals that deviate from the group distribution characteristics. The recommended action generator integrates knowledge graph and case library resources to automatically match pre-set solution templates based on anomaly types and dynamically generate adjustment recommendations based on real-time environmental parameters. The trigger time node calculation module utilizes queuing theory models to comprehensively consider the urgency of the issue, resource availability, and repair cycle to determine the optimal intervention time. The intervention record analysis module in step S4 establishes a multidimensional evaluation index system to quantitatively assess the effectiveness of improvement plans based on three dimensions: response speed, execution efficiency, and sustainability of results. The parameter space optimization algorithm utilizes a Bayesian optimization framework, using historical intervention data as prior knowledge and using Gaussian processes to model the nonlinear relationship between the parameter space and evaluation results. The model update strategy utilizes a shadow run mode. Newly generated weight parameters are first simulated and validated in a parallel system, and then gradually replaced with existing parameters after comparative performance testing. The system's self-check module incorporates an exception circuit breaker mechanism. If a cluster of deviations in evaluation results are detected, the system automatically rolls back to the previous stable version and triggers a manual review process. This approach is suitable for team performance management in knowledge-intensive enterprises, and is particularly well-suited for the dynamic assessment needs of project-based organizational structures. In a market environment where strategic objectives are frequently adjusted, the system maintains high consistency between evaluation criteria and organizational strategy; and in response to sudden business transformations, it can rapidly reconfigure evaluation model parameters. The implementation demonstrates three core advantages: evaluation timeliness, enabling a shift from monthly assessments to real-time monitoring; indicator adaptability, supporting automatic reconfiguration of the evaluation system following strategic objective adjustments; and management foresight, transforming post-evaluation into pre-emptive intervention through an anomaly warning mechanism. Compared to traditional methods, the system significantly improves indicator relevance, timely feedback, and strategic accuracy.

[0033] The scientific research performance information analysis and evaluation system and its processing method involved in this embodiment solve the problem by constructing a dynamic weight allocation module and a real-time feedback intervention mechanism: the data acquisition module integrates multi-source heterogeneous data such as attendance records, task status, text interaction and sensor behavior, and generates a first signal with a timestamp after standardization; the weight allocation module uses an online learning algorithm to analyze the strategic goal priority, dynamically constructs a weight parameter space, reduces the impact of historical data through a preset attenuation factor, and generates a second signal for real-time weight mapping; the real-time feedback module constructs a multi-dimensional evaluation matrix based on the weight relationship, monitors indicator deviations and triggers a third signal containing abnormal codes and improvement measures; the execution engine module generates personalized plans through similarity matching, aggregates intervention records to form a fourth signal, and feeds back to the weight allocation module, forming a closed-loop system of dynamic weight adjustment and execution optimization, thereby realizing strategic-oriented precise evaluation and automated intervention.

[0034] The scientific research performance information analysis and evaluation system and its operation and processing method involved in this embodiment also solve the problems of traditional performance evaluation systems relying on static weights, having a single data dimension and lacking real-time dynamic adjustment capabilities.

[0035] This embodiment also provides a method for using the scientific research performance information analysis and evaluation system, which specifically includes the following steps during the scientific research performance analysis and evaluation process: (1) Users need to use their account information to access the system when logging in. Users cannot log in without a set account. If the login fails due to an incorrect username or password, the system will pop up a prompt dialog box. Re-enter the correct information to successfully log in and enter the main page. The main interface is designed to be friendly, simple and easy to operate. (2) The system navigation bar contains four major functional modules. Users can operate in different modules. Each module specifically includes the following operations: View: Click the "View" button in the operation bar to view detailed information; Modify: Click the "Modify" button to enter the editing page, adjust the options and submit to save, the system will prompt that the operation is successful; Delete: Click the "Delete" button and confirm the prompt. The system will prompt that the deletion is successful and the data is removed from the list. Add: Click the "Add" button to fill in basic information. After submitting, the system will display "Operation Successful". Return to the list page to view the new data. Search: Enter keywords in the search bar in the upper right corner of the interface, and the system will filter and display matching results; The four major functional modules are: (2.1) Employee Performance Entry: supports the entry of personal performance, attendance data, project results, rewards and punishments; each entry sub-module requires seven parameters to be filled in, specifically: Personal performance entry: title, employee name, performance month, performance type, performance amount, performance description, and review status; Attendance data entry: title, employee name, attendance date, attendance type, attendance duration, attendance remarks, and review status; Project achievement entry: title, project name, achievement type, achievement description, achievement completion time, achievement status, and review status; Reward and punishment information entry: title, employee name, reward and punishment type, reward and punishment reason, reward and punishment amount, reward and punishment date, and review status; (2.2) Department Performance Summary: Provides a summary of each position's performance, team project performance, and monthly and quarterly performance. Data visualization charts intuitively display complex data, helping users quickly identify key information and trends. Traditional data reports contain numbers and text that are difficult to quickly understand, but charts, combined with graphics, colors, and layout, make data relationships clear at a glance, improving the efficiency of information transmission.

Claims

1. A scientific research performance information analysis and evaluation system, characterized by: The data acquisition module generates a first signal containing a timestamp; the weight allocation module receives the first signal and parses the current strategic goal priority identifier and historical performance evaluation records contained therein, and constructs a weight parameter space that dynamically matches the strategic goal priority through an online learning algorithm, wherein each weight coefficient in the weight parameter space automatically reduces the influence of historical data over time according to a preset attenuation factor, thereby generating a second signal containing a real-time weight mapping relationship; The real-time feedback module constructs a multidimensional performance evaluation matrix based on the weight mapping relationship in the second signal, and generates a third signal containing an abnormality location code, a list of recommended measures and a trigger time node; the execution engine module receives the third signal and parses the abnormality location code contained therein, matches the recommended measures list with the employee capability archive for similarity to generate a personalized improvement plan, and at the same time aggregates all intervention records in the current cycle to form a fourth signal, and the fourth signal is fed back to the weight allocation module to drive the iterative optimization of the weight parameter space.

2. A scientific research performance information analysis and evaluation system according to claim 1, characterized in that: The data acquisition module includes a multi-source data fusion engine, which converts heterogeneous data formats for structured attendance data from an enterprise resource planning system, semi-structured task logs from an office automation system, unstructured text interaction content collected by employee terminal devices, and time-series behavior data generated by wearable device sensors, establishes a cross-modal data association relationship map, and maps key indicators representing work engagement in different data sources to a unified vector space through a feature extractor; after completing data alignment, the multi-source data fusion engine injects a timestamp tag to form the first signal, wherein the unstructured text interaction content is divided into dialogue topic paragraphs and participant identifiers are attached by a semantic segmentation unit; the sensor behavior data generates a periodic behavior pattern feature vector through a sliding window statistical method.

3. A scientific research performance information analysis and evaluation system according to claim 2, characterized in that: The system further includes a natural language processing unit, which performs dependency syntax analysis and sentiment polarity calculation on the text interaction content, extracts the frequency distribution matrix of keywords related to task allocation in the dialogue segment, identifies the innovative suggestions put forward by the speaker and constructs a contribution intensity index; the natural language processing unit captures the contextual association relationship in the long text through the attention mechanism, generates an additional feature vector containing a collaboration score and injects it into the first signal set, wherein the collaboration score is weighted according to the number of times the topic is actively raised and the frequency of the solution being adopted in the dialogue round; The sentiment polarity calculation results are used to correct the semantic weight of text interaction content.

4. A scientific research performance information analysis and evaluation system according to claim 1, characterized in that: The weight allocation module also includes a dynamic adjustment unit, through which the attenuation factor preset in the weight allocation module is nonlinearly changed; the dynamic adjustment unit receives a target change notification signal from the enterprise strategic management system, and when it detects that the strategic priority has switched across dimensions, it initiates an attenuation factor recalibration process, and calculates a transition curve equation between the new and old weights based on the correlation matrix before and after the strategic target switch; The attenuation factor recalibration process includes the gradual attenuation of the influence of historical weights and the introduction of exponential smoothing of new strategic weights; after completing the parameter update, the dynamic adjustment unit generates a weight migration trajectory report and writes it into the system audit log.

5. A scientific research performance information analysis and evaluation system according to claim 1, characterized in that: The online learning algorithm in the weight allocation module adopts a two-layer optimization framework, wherein the first-layer optimizer is configured to construct an initial weight parameter space based on the real-time strategic goal priority, and the second-layer optimizer is configured to receive the intervention effect feedback data carried in the fourth signal from the execution engine module, and adjust the loss function constraints of the first-layer optimizer through the backpropagation mechanism; the two-layer optimization framework realizes the coordinated optimization of strategic guidance and execution feedback through alternating iterations, wherein the second-layer optimizer calculates the elasticity coefficient of the intervention measures on the performance evaluation indicators during each iteration, and injects the coefficient as a regularization term into the objective function of the first-layer optimizer.

6. A scientific research performance information analysis and evaluation system according to claim 1, characterized in that: The multi-dimensional performance evaluation matrix in the real-time feedback module includes a longitudinal time dimension and a transverse capability dimension. The longitudinal time dimension divides the evaluation period by a sliding time window and establishes a historical performance trend regression model. The horizontal capability dimension constructs a capability item association rule library based on the job description. The deviation detection unit is configured to calculate the standardized residual value of each capability item of the individual indicator in the horizontal dimension. When the residual value of the same capability item exceeds two standard deviations of the team benchmark value in three consecutive evaluation cycles, the progressive intervention strategy is activated. The progressive intervention strategy includes three levels of progressive measures: early warning notification, targeted training suggestions, and job adaptability adjustment plan.

7. A scientific research performance information analysis and evaluation system according to claim 1, characterized in that: The real-time feedback module further includes a situational awareness unit, which is configured to receive industry volatility index and market competition intensity data from the enterprise's external environment monitoring system, construct a dynamic threshold adjustment model, and automatically lower the deviation detection threshold sensitivity when it is detected that the external environment change rate exceeds a preset critical value; the situational awareness unit extracts the periodic fluctuation characteristics in the external environment data through a convolutional neural network, and encodes it as an environmental pressure coefficient and injects it into the weight correction term of the multidimensional performance evaluation matrix, where the elastic relationship between the environmental pressure coefficient and the team baseline value is modeled through a multiple regression equation.

8. A scientific research performance information analysis and evaluation system according to claim 1, characterized in that: The similarity matching process in the execution engine module adopts a multimodal embedding model, which is configured to map the text descriptions of skill certificates, project experience structured data and historical training records in the employee capability archive to a unified semantic space to generate employee capability feature vectors. At the same time, the intervention measure text in the recommended measures list is sentence-vector encoded, the cosine similarity between the two is calculated, and the job competency model is introduced as a constraint condition; when generating personalized improvement plans, the multimodal embedding model retains the top K optimal matching results and adds a feasibility assessment score, which is dynamically adjusted according to the ratio of the implementation cost matrix to the expected benefit matrix.

9. A scientific research performance information analysis and evaluation system according to claim 1, characterized in that: The execution engine module further includes an intervention effect tracking unit, which is configured to start an effect monitoring process after sending a personalized improvement plan, calculate the effectiveness coefficient of the measure by comparing the change rate of the performance evaluation indicator of the target capability item before and after the intervention, and automatically trigger the alternative plan generation mechanism when it is detected that the effectiveness coefficient is lower than a preset level; The intervention effect tracking unit inputs the effectiveness coefficient sequence within each period into the time series prediction model to generate the effect attenuation curve for the next three evaluation periods and optimizes the parameter settings of the similarity matching algorithm accordingly.

10. A method for analyzing and evaluating scientific research performance information, characterized in that: The specific process steps of performing information analysis and processing by means of the computer system according to claim 1 include: S1: Information acquisition and vectorization processing: First, obtain multi-source heterogeneous data including attendance records, text interaction content, and sensor behavior data, then perform standardization cleaning and feature vectorization processing on the multi-source heterogeneous data to form a first signal with a timestamp mark; S2: Constructing a weight parameter space: Receive the first signal and parse the current strategic goal priority identifier and historical performance evaluation records contained therein. Use an online learning algorithm to construct a weight parameter space that dynamically matches the strategic goal priority. Each weight coefficient in the weight parameter space automatically reduces the influence of historical data over time according to a preset attenuation factor, generating a second signal containing a real-time weight mapping relationship. S3: Constructing a performance evaluation matrix: Constructing a multi-dimensional performance evaluation matrix based on the weight mapping relationship in the second signal, and generating a third signal containing an anomaly location code, a list of recommended measures, and a trigger time node; S4: Optimize weight parameter space: Receive the third signal and parse the abnormal location code contained in it, perform similarity matching between the list of recommended measures and the employee capability archive to generate a personalized improvement plan, and at the same time aggregate all intervention records in the current cycle to form the fourth signal to drive the iterative optimization of the weight parameter space.

Images