Method and device for generating capability profile data based on multi-source heterogeneous data fusion

Abstract

The application relates to a capability portrait data generation method and device based on multi-source heterogeneous data fusion, computer equipment, a computer readable storage medium and a computer program product, which can be used in the technical field of big data. The method comprises the following steps: obtaining original data of a target object from a plurality of heterogeneous data sources; the original data comprises structured data and unstructured data; inputting the unstructured data into a pre-trained entity recognition model for entity recognition processing to obtain target entity data in the unstructured data; performing data fusion processing on the target entity data and the structured data to obtain fusion feature data of the target object; inputting the fusion feature data into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions; and generating capability portrait data of the target object according to the capability prediction data. The method can improve the accuracy of capability portrait data generation.

Description

Technical Field

[0001] This application relates to the field of big data technology, and in particular to a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for generating capability profile data based on the fusion of multi-source heterogeneous data. Background Technology

[0002] With the deepening of digital transformation, capability profiling, as a technical means to comprehensively depict the capability status of a target object, has been widely used.

[0003] However, in practical applications, the capability-related data of the target object is often stored in different business systems, and the data types are complex and diverse. The relevant capability profile data generation schemes are difficult to effectively process the above-mentioned complex data, resulting in low accuracy of capability profile data generation. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product for generating capability profile data based on multi-source heterogeneous data fusion, which can improve the accuracy of capability profile data generation and address the aforementioned technical problems.

[0005] Firstly, this application provides a method for generating capability profile data based on the fusion of multi-source heterogeneous data. The method includes:

[0006] The raw data of the target object is obtained from multiple heterogeneous data sources; the raw data includes structured data and unstructured data.

[0007] The unstructured data is input into a pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data;

[0008] The target entity data and the structured data are fused together to obtain the fused feature data of the target object.

[0009] The fused feature data is input into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions.

[0010] Based on the capability prediction data, capability profile data of the target object is generated.

[0011] In one embodiment, the step of inputting the fused feature data into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions includes:

[0012] The static feature data in the fused feature data is input into the static feature processing network in the capability prediction model for capability prediction processing, thereby obtaining the static feature vector of the target object;

[0013] The temporal feature data in the fused feature data is input into the temporal feature processing network in the capability prediction model for capability prediction processing, thereby obtaining the temporal feature vector of the target object;

[0014] The static feature vector and the temporal feature vector are dynamically fused through the attention mechanism layer in the capability prediction model to obtain the fused feature vector of the target object.

[0015] The capability prediction data is determined based on the fused feature vector.

[0016] In one embodiment, before determining the capability prediction data based on the fused feature vector, the method further includes:

[0017] Obtain information about the current application scenario;

[0018] Based on the current application scenario information, determine the weight information corresponding to each dimension;

[0019] The step of determining the capability prediction data based on the fused feature vector includes:

[0020] The capability prediction data is determined based on the fused feature vector and the weight information.

[0021] In one embodiment, the multiple dimensions include experience dimension, professional dimension, achievement dimension and potential dimension;

[0022] The step of generating capability profile data for the target object based on the capability prediction data includes:

[0023] Based on the target object's predicted capabilities under the experience dimension, professional dimension, achievement dimension, and potential dimension, generate the target object's capability distribution map data;

[0024] Based on the capability distribution map data, the capability profile data is determined.

[0025] In one embodiment, after generating the capability profile data of the target object based on the capability prediction data, the method further includes:

[0026] Obtain target requirement information;

[0027] The target demand information is subjected to feature extraction processing to obtain a demand feature vector;

[0028] Based on the demand feature vector and the capability profile data, determine the degree of matching between the target object and the target demand information;

[0029] Based on the matching degree, an object recommendation result corresponding to the target demand information is generated.

[0030] In one embodiment, the capability prediction model is trained in the following manner:

[0031] Acquire historical feature data and real capability data of multiple sample objects;

[0032] Based on the historical feature data and the actual capability data, the capability prediction model to be trained is iteratively trained to obtain the capability prediction model.

[0033] Secondly, this application also provides a capability profile data generation device based on multi-source heterogeneous data fusion. The device includes:

[0034] The data acquisition module is used to acquire raw data of the target object from multiple heterogeneous data sources; the raw data includes structured data and unstructured data.

[0035] The first input module is used to input the unstructured data into a pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data;

[0036] The data fusion module is used to perform data fusion processing on the target entity data and the structured data to obtain the fused feature data of the target object;

[0037] The second input module is used to input the fused feature data into a pre-trained capability prediction model for capability prediction processing, so as to obtain capability prediction data of the target object in multiple dimensions.

[0038] The data generation module is used to generate capability profile data of the target object based on the capability prediction data.

[0039] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:

[0040] The raw data of the target object is obtained from multiple heterogeneous data sources; the raw data includes structured data and unstructured data.

[0041] The unstructured data is input into a pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data;

[0042] The target entity data and the structured data are fused together to obtain the fused feature data of the target object.

[0043] The fused feature data is input into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions.

[0044] Based on the capability prediction data, capability profile data of the target object is generated.

[0045] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:

[0046] The raw data of the target object is obtained from multiple heterogeneous data sources; the raw data includes structured data and unstructured data.

[0047] The unstructured data is input into a pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data;

[0048] The target entity data and the structured data are fused together to obtain the fused feature data of the target object.

[0049] The fused feature data is input into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions.

[0050] Based on the capability prediction data, capability profile data of the target object is generated.

[0051] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, performs the following steps:

[0052] The raw data of the target object is obtained from multiple heterogeneous data sources; the raw data includes structured data and unstructured data.

[0053] The unstructured data is input into a pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data;

[0054] The target entity data and the structured data are fused together to obtain the fused feature data of the target object.

[0055] The fused feature data is input into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions.

[0056] Based on the capability prediction data, capability profile data of the target object is generated.

[0057] The aforementioned capability profile data generation method, apparatus, computer equipment, computer-readable storage medium, and computer program product based on multi-source heterogeneous data fusion, by acquiring raw data containing both structured and unstructured data of the target object from multiple heterogeneous data sources, helps to break down data silos and fully collect multi-dimensional information of the target object across different business platforms, thereby improving the comprehensiveness of subsequent data fusion. Furthermore, by inputting unstructured data into a pre-trained entity recognition model for entity recognition processing to obtain target entity data, it facilitates in-depth mining of free text data that is difficult to quantify using traditional methods, extracting entity information with clear semantics, thus improving the utilization rate of unstructured data and the accuracy of information extraction. Accuracy; By fusing target entity data and structured data to obtain fused feature data, it is beneficial to associate and integrate multi-dimensional information from different data sources, construct a full-dimensional feature representation of the target object, and thus improve the data foundation integrity of capability assessment; By inputting the fused feature data into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data in multiple dimensions, it is beneficial to explore the deep correlation between the fused feature data, and thus improve the accuracy of capability prediction data; By generating capability profile data of the target object based on the capability prediction data, it is beneficial to present the comprehensive capability characteristics of the target object in a multi-dimensional quantitative manner, and thus improve the accuracy of capability profile data generation. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart illustrating a capability profile data generation method based on multi-source heterogeneous data fusion in one embodiment.

[0060] Figure 2 This is a flowchart illustrating the steps for determining capability prediction data in one embodiment;

[0061] Figure 3This is a flowchart illustrating the steps for determining capability prediction data in another embodiment;

[0062] Figure 4 This is a structural block diagram of a capability profile data generation device based on multi-source heterogeneous data fusion in one embodiment.

[0063] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0066] In one exemplary embodiment, such as Figure 1 As shown, a method for generating capability profile data based on multi-source heterogeneous data fusion is provided. This embodiment illustrates the application of this method to a terminal; it is understood that this method can also be applied to a server, and to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, tablets, etc.; the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. In this embodiment, the method includes the following steps:

[0067] Step S101: Obtain the original data of the target object from multiple heterogeneous data sources; the original data includes structured data and unstructured data.

[0068] Step S102: Input the unstructured data into the pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data.

[0069] Step S103: Perform data fusion processing on the target entity data and structured data to obtain the fused feature data of the target object;

[0070] Step S104: Input the fused feature data into the pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions;

[0071] Step S105: Generate capability profile data for the target object based on the capability prediction data.

[0072] Among them, capability profile data can refer to a data set used to describe the comprehensive capability characteristics distribution of a target object across multiple capability dimensions. Capability profile data can intuitively present the target object's strengths and areas for improvement.

[0073] Among them, multiple heterogeneous data sources can refer to multiple data source platforms that use different data storage formats or different data organization structures, such as independent business platforms such as office automation systems, project management systems, scientific research result databases, human resource systems, and financial systems.

[0074] The target audience can refer to specific individuals who require capability assessment and profile building, such as R&D personnel in the field of technological innovation.

[0075] Raw data can refer to initial data obtained directly from multiple heterogeneous data sources without fusion processing. Raw data includes structured data and unstructured data.

[0076] Structured data can refer to data stored according to predefined field formats and data models, such as basic personnel information, education experience, work experience, performance appraisal scores, skill certificate records, project funding data, etc., organized in tabular form.

[0077] Unstructured data can refer to free text data that does not have a fixed field format, such as project description reports, technical documents, creative proposals, honorary certificate texts, and other text-based data.

[0078] The pre-trained entity recognition model can refer to a deep learning model built on a pre-trained language model, used to automatically identify and extract specific types of entity information from free text. This entity recognition model has been trained with labeled corpora before use.

[0079] Entity recognition processing can refer to the process of using entity recognition models to parse text content in unstructured data, automatically locate and extract named entities with specific semantic categories from the text.

[0080] The target entity data can refer to entity information with a specific semantic category extracted from unstructured data through entity recognition processing, such as project name, technical role, innovative achievement, knowledge number, award level, and other entity information.

[0081] Data fusion processing can refer to the process of associating, aligning, and integrating target entity data and structured data according to a unified master data model. For example, it can aggregate and associate multi-source data based on personnel identification and timestamps to generate a complete data record covering multiple dimensions.

[0082] Among them, fused feature data can refer to a data set that has been processed by data fusion and covers comprehensive feature information of the target object in multiple dimensions. Fused feature data includes, but is not limited to, feature information in multiple dimensions such as basic information, education experience, project experience, scientific research results, innovation and creativity, and performance evaluation.

[0083] Among them, the pre-trained capability prediction model can refer to a machine learning model based on neural networks that is used to predict the capability level of a target object in multiple dimensions based on fused feature data. This capability prediction model has been trained on a labeled dataset constructed from historical data and result data before it is used.

[0084] Capability prediction processing can refer to the process of taking fused feature data as input and using a capability prediction model to quantitatively evaluate and predict the capability level of a target object in various dimensions.

[0085] Among them, multiple dimensions can refer to multiple assessment aspects used to evaluate the comprehensive capabilities of the target object, such as experience dimension, professional dimension, achievement dimension and potential dimension. The experience dimension is used to assess the depth of educational background and project experience, the professional dimension is used to assess the competence of technical positions and the value of suggestions for implementation, the achievement dimension is used to assess the value of technical awards and intellectual property transformation, and the potential dimension is used to assess innovation leadership and technical insight.

[0086] Among them, capability prediction data can refer to the data output by the capability prediction model, which represents the quantitative score and potential prediction value of the target object's capabilities in multiple dimensions.

[0087] Optionally, the terminal obtains the target object's raw data from multiple heterogeneous data sources. The raw data includes structured and unstructured data. Structured data includes tabular data such as basic personnel information, educational background, work experience, project participation records, performance evaluation scores, and skill certificate records. Unstructured data includes free text data such as project description reports, technical documents, creative proposals, and honorary certificate texts. The terminal inputs the unstructured data into a pre-trained entity recognition model for entity recognition processing, obtaining the target entity data from the unstructured data. The target entity data includes entity information extracted from the text content of the unstructured data, such as project name, technical role, innovative achievements, intellectual property number, and award level. The terminal performs data fusion processing on the target entity data and structured data. Based on a unified master data model, the terminal aligns the target entity data with the structured data according to personnel identifiers and timestamps, obtaining the target object's fused feature data. The fused feature data covers comprehensive feature information of the target object across multiple dimensions, including basic information, educational background, project experience, research achievements, innovative ideas, and performance evaluations. The terminal inputs the fused feature data into a pre-trained capability prediction model for capability prediction processing, obtaining capability prediction data for the target object across multiple dimensions, including experience, expertise, achievements, and potential. Based on the capability prediction data, the terminal generates capability profile data for the target object, which visually presents the distribution of the target object's capability characteristics across multiple dimensions.

[0088] For example, the terminal connects to multiple heterogeneous data sources, such as office automation systems, project management systems, scientific research result databases, and human resource systems, through data acquisition tools, automatically retrieving the raw data of the target object from each source. For structured data in the raw data, the terminal directly reads and stores it according to predefined field formats. For instance, it retrieves the target object's educational background and performance evaluation scores from the human resource system, and project participation records and project funding data from the project management system. For unstructured data in the raw data, the terminal inputs it into a pre-trained entity recognition model for entity recognition processing. Based on a pre-trained language model architecture, the entity recognition model performs deep analysis of the text content in unstructured data such as project description reports and technical documents, automatically identifying and extracting target entity data. For example, it extracts the project name and technical role from a project description report, and innovative achievements and intellectual property numbers from a technical document. The terminal performs data fusion processing on target entity data and structured data. Specifically, based on a unified talent master data model, the terminal uses the personnel identifier of the target object and the corresponding timestamp as the association key to align and aggregate target entity data and structured data from different heterogeneous data sources, generating fused feature data of the target object. This fused feature data constitutes a full-dimensional data record covering multiple dimensions such as basic information, education experience, work experience, project experience, scientific research achievements, innovation and creativity, performance appraisal, and skills certificates. The terminal inputs the fused feature data into a pre-trained capability prediction model for capability prediction processing. The capability prediction model adopts a hybrid neural network architecture, processing static features in the fused feature data through a static feature processing sub-network, processing temporal features in the fused feature data through a temporal feature processing sub-network, and dynamically fusing the processing results of the two types of features through an attention mechanism. The output is capability prediction data of the target object in the experience dimension, professional dimension, achievement dimension, and potential dimension. The capability prediction data includes the target object's quantitative capability score and potential prediction value in each dimension. Based on the capability prediction data, the terminal generates capability profile data for the target object. The capability profile data presents the target object's strengths and areas for improvement in the form of multi-dimensional capability feature distribution across experience, expertise, achievements, and potential dimensions.

[0089] For example, after generating a capability profile of a target object based on capability prediction data, corresponding task allocation can be performed based on this profile. For instance, a high-difficulty task (e.g., a task with a difficulty coefficient greater than a preset threshold) can be assigned to an object whose capability profile data exceeds a preset capability threshold (i.e., an object with strong capabilities), thereby improving task processing efficiency. For example, if a target object's capability profile data exceeds a preset capability threshold, a preset task (i.e., a high-difficulty task) can be assigned to that target object, which will then process the preset task.

[0090] The aforementioned capability profile data generation method based on multi-source heterogeneous data fusion, by acquiring raw data containing both structured and unstructured data of the target object from multiple heterogeneous data sources, helps to break down data silos and fully collect multi-dimensional information of the target object across different business platforms, thereby improving the comprehensiveness of subsequent data fusion. By inputting unstructured data into a pre-trained entity recognition model for entity recognition processing to obtain target entity data, it facilitates in-depth mining of free text data that is difficult to quantify using traditional methods, extracting entity information with clear semantics, thus improving the utilization rate of unstructured data and the accuracy of information extraction. Furthermore, by analyzing the target entity data and... Structured data fusion processing to obtain fused feature data facilitates the association and integration of multi-dimensional information from different data sources, constructing a full-dimensional feature representation of the target object, thereby improving the data foundation integrity for capability assessment. Inputting the fused feature data into a pre-trained capability prediction model for capability prediction processing, resulting in multi-dimensional capability prediction data, helps to uncover deep correlations between the fused feature data, thus improving the accuracy of capability prediction data. Generating capability profile data for the target object based on the capability prediction data allows for the presentation of the target object's comprehensive capability characteristics in a multi-dimensional quantitative manner, thereby improving the accuracy of capability profile data generation.

[0091] In one exemplary embodiment, reference is made to Figure 2 The fused feature data is input into a pre-trained capability prediction model for capability prediction processing, resulting in capability prediction data for the target object across multiple dimensions, including:

[0092] Step S201: Input the static feature data from the fused feature data into the static feature processing network in the capability prediction model for capability prediction processing to obtain the static feature vector of the target object;

[0093] Step S202: Input the temporal feature data from the fused feature data into the temporal feature processing network in the capability prediction model for capability prediction processing to obtain the temporal feature vector of the target object;

[0094] Step S203: Through the attention mechanism layer in the capability prediction model, the static feature vector and the temporal feature vector are dynamically fused to obtain the fused feature vector of the target object.

[0095] Step S204: Determine the capability prediction data based on the fused feature vector.

[0096] Among them, static feature data can be feature data in the fused feature data that does not change over time, such as fixed attribute features such as the target object's educational background, professional title information, and registration certificate information.

[0097] Among them, the static feature processing network can be a sub-network in the capability prediction model used to process static feature data. This static feature processing network adopts a multilayer perceptron structure and performs nonlinear mapping processing on the static feature data through multiple fully connected layers.

[0098] Among them, the static feature vector can be a fixed-dimensional numerical vector output by static feature data after being processed by a static feature processing network, which is used to characterize the static capability features of the target object.

[0099] Among them, time-series feature data can be feature data that changes dynamically over time in the fused feature data, such as the target object's project participation experience, scientific research achievements, performance evaluation scores, and other feature data arranged in a time series.

[0100] Among them, the temporal feature processing network can be a sub-network in the capability prediction model used to process temporal feature data. This temporal feature processing network adopts a gated recurrent unit structure, which can capture the time dependence and dynamic change trend in the temporal feature data.

[0101] Among them, the temporal feature vector can be a fixed-dimensional numerical vector output after the temporal feature data is processed by the temporal feature processing network, which is used to characterize the dynamic capability growth characteristics of the target object.

[0102] The attention mechanism layer can be a network layer in the capability prediction model used to dynamically weight and fuse static feature vectors and temporal feature vectors. This attention mechanism layer can adaptively allocate attention weights according to the importance of different features.

[0103] The dynamic fusion process can be a process in which the static feature vector and the temporal feature vector are adaptively weighted and combined according to their respective importance through an attention mechanism layer.

[0104] The fused feature vector can be a comprehensive numerical vector obtained by dynamically fusing static feature vectors and time-series feature vectors. The fused feature vector contains both the static capability features and the dynamic capability growth features of the target object.

[0105] Optionally, the terminal separates static feature data and temporal feature data from the fused feature data. The static feature data is input into the static feature processing network in the capability prediction model for processing. This network performs non-linear mapping on the static feature data through multiple fully connected layers to obtain the static feature vector of the target object. The terminal then inputs the temporal feature data into the temporal feature processing network in the capability prediction model for processing. This network models the time-series information in the temporal feature data through gated recurrent units to obtain the temporal feature vector of the target object. The terminal then uses the attention mechanism layer in the capability prediction model to dynamically weight and fuse the static and temporal feature vectors based on their feature importance, obtaining the fused feature vector of the target object. Finally, the terminal uses the fused feature vector to perform mapping processing through the output layer of the capability prediction model to determine the capability prediction data of the target object across multiple dimensions.

[0106] The technical solution provided in this embodiment, by inputting the static feature data and temporal feature data from the fused feature data into the static feature processing network and the temporal feature processing network respectively for processing, is beneficial to capture the fixed attribute features and dynamic growth features of the target object respectively. By dynamically fusing the static feature vector and the temporal feature vector through the attention mechanism layer, it is beneficial to adaptively allocate weights according to the importance of different features, thereby improving the accuracy and comprehensiveness of capability prediction data.

[0107] In one exemplary embodiment, reference is made to Figure 3 Before determining the capability prediction data based on the fused feature vector, the following steps are also included:

[0108] Step S301: Obtain current application scenario information;

[0109] Step S302: Determine the weight information corresponding to each dimension based on the current application scenario information;

[0110] Based on the fused feature vectors, capability prediction data is determined, including:

[0111] Step S303: Determine the capability prediction data based on the fused feature vector and weight information.

[0112] The current application scenario information can describe the specific business scenario for which the current capability assessment is being conducted. For example, the current application scenario information could be a scenario for selecting young talents, evaluating leading talents, selecting annual innovative talents, project initiation, or professional title evaluation.

[0113] The weighting information for each dimension can be determined based on the proportion of each evaluation dimension according to the current application scenario. Different current application scenarios correspond to different weighting information for each dimension. For example, when the current application scenario is for selecting young talents, the weighting information for the potential dimension can be 40%, and the weighting information for the achievement dimension can be 20%; when the current application scenario is for evaluating leading talents, the weighting information for the achievement dimension can be 50%, and the weighting information for the potential dimension can be 15%.

[0114] Optionally, before determining the capability prediction data based on the fused feature vector, the terminal obtains the current application scenario information, which indicates the specific business scenario for which the capability assessment is being conducted. Based on the current application scenario information, the terminal constructs a judgment matrix using the analytic hierarchy process (AHP) to determine the weight information corresponding to each dimension. This weight information is used to adjust the proportion of each assessment dimension in the capability prediction data. The terminal then performs weighted processing on the feature components of each dimension in the fused feature vector based on the fused feature vector and the weight information corresponding to each dimension, obtaining the capability prediction data of the target object across multiple dimensions.

[0115] The technical solution provided in this embodiment obtains current application scenario information and determines the weight information corresponding to each dimension based on the current application scenario information. This is beneficial for dynamically adjusting the weight ratio of each dimension according to the actual evaluation needs of different business scenarios. By determining the capability prediction data based on the fused feature vector and weight information, it is beneficial for improving the matching degree between the capability prediction data and the current application scenario, thereby improving the relevance and applicability of the capability prediction data.

[0116] In an exemplary embodiment, multiple dimensions include experience dimension, professional dimension, achievement dimension and potential dimension; based on the capability prediction data, generating capability profile data for the target object includes: generating capability distribution map data for the target object based on the capability prediction data of the target object under the experience dimension, professional dimension, achievement dimension and potential dimension; and determining capability profile data based on the capability distribution map data.

[0117] Among them, the experience dimension can be used to assess the matching degree of the target object's educational background and major, the depth of research project experience, and the breakthroughs in technical research.

[0118] Among them, the professional dimension can be used to evaluate the target object's key technical job competency, the value of rationalization proposals, the benchmark nature of operating procedures, the radiation effect of talent training, the authority of registration certificates, and the stability of recent performance appraisals.

[0119] Among them, the achievement dimension can be an assessment dimension used to evaluate the target object's level of technical awards, academic influence, intellectual property transformation value, and industry recognition of standards and norms.

[0120] Among them, the potential dimension can be an assessment dimension used to evaluate the target object's innovation leadership, technological insight, systems thinking ability, and resilience in overcoming difficulties.

[0121] Among them, the capability distribution map data can be visualized graphical data generated based on the capability prediction data of the target object under the experience dimension, professional dimension, achievement dimension and potential dimension. The capability distribution map data displays the distribution of the target object's capability strength and weakness under each dimension in the form of a radar chart.

[0122] Optionally, the terminal acquires predicted capability data for the target object across four dimensions: experience, professional background, achievements, and potential. The experience dimension assesses the target object's educational background and project experience; the professional dimension assesses their job competence and the practical value of their recommendations; the achievements dimension assesses their technical awards and intellectual property commercialization value; and the potential dimension assesses their innovation leadership and technological insight. Based on this predicted capability data, the terminal generates a capability distribution map of the target object, using the predicted capability data for each dimension as values ​​for each axis of a radar chart. Based on this capability distribution map data, the terminal determines the target object's capability profile data, which includes the capability distribution map data and annotations of the target object's strengths and areas for improvement across each dimension.

[0123] The technical solution provided in this embodiment, by subdividing multiple dimensions into experience dimension, professional dimension, achievement dimension and potential dimension, is conducive to comprehensively covering the explicit capabilities and implicit potential of the target object from multiple perspectives. By generating capability distribution map data based on the capability prediction data of the target object under each dimension and determining capability profile data, it is beneficial to present the capability characteristic distribution of the target object in a visual way, thereby improving the comprehensiveness of capability profile data.

[0124] In an exemplary embodiment, after generating capability profile data of the target object based on capability prediction data, the method further includes: obtaining target demand information; performing feature extraction processing on the target demand information to obtain a demand feature vector; determining the matching degree between the target object and the target demand information based on the demand feature vector and the capability profile data; and generating object recommendation results corresponding to the target demand information based on the matching degree.

[0125] The target requirement information can be a description of specific project requirements or job requirements published in the business scenario, such as skill tag requirements, experience requirements, and role requirements for technical breakthrough projects.

[0126] The demand feature vector can be a numerical vector obtained after feature extraction processing of the target demand information. The demand feature vector is used to characterize the skill and experience features required in the target demand information.

[0127] Among them, the matching degree can be a numerical indicator that measures the degree of matching between the capability profile data of the target object and the target demand information. The matching degree can be determined by calculating the cosine similarity between the demand feature vector and the profile feature vector corresponding to the capability profile data.

[0128] The object recommendation result can be recommendation information generated based on the matching degree, indicating whether the target object is suitable for the target needs. The object recommendation result can include a list of recommended candidates sorted by matching degree, as well as the matching degree and recommendation reason for each recommended candidate.

[0129] Optionally, after generating the capability profile data of the target object, the terminal obtains the target requirement information, which is the project requirement or job requirement published in the business scenario. The terminal performs feature extraction processing on the target requirement information, extracting skill tags and experience requirements from the target requirement information, and encodes the extraction results into a requirement feature vector. Based on the requirement feature vector and the capability profile data of the target object, the terminal calculates the similarity between the requirement feature vector and the corresponding profile feature vector of the capability profile data to determine the matching degree between the target object and the target requirement information. Based on the matching degree, the terminal sorts the matching degrees and generates object recommendation results corresponding to the target requirement information. The object recommendation results include a list of recommended candidates sorted from high to low matching degree.

[0130] The technical solution provided in this embodiment obtains target demand information and performs feature extraction processing on the target demand information to obtain a demand feature vector. This helps to transform business demands into calculable numerical features. By determining the matching degree based on the demand feature vector and capability profile data and generating object recommendation results, it is beneficial to achieve accurate matching between target objects and business demands, thereby improving the accuracy of object recommendations and the efficiency of business decision-making.

[0131] In an exemplary embodiment, the capability prediction model is trained by: acquiring historical feature data and actual capability data of multiple sample objects; and iteratively training the capability prediction model to be trained based on the historical feature data and actual capability data to obtain the capability prediction model.

[0132] Among them, multiple sample objects can be historical individuals with known ability levels used to train the ability prediction model. Each sample object has corresponding historical feature data and actual ability data.

[0133] Historical feature data can be the fusion feature data of multiple sample objects within a historical time period. Historical feature data includes feature information such as the four-dimensional indicator vector of each sample object, historical project success rate, and innovative output.

[0134] Among them, real capability data can be real capability score data of multiple sample objects in various dimensions that have been manually labeled or actually verified. Real capability data is used as supervision labels in the training process.

[0135] Among them, the capability prediction model to be trained can be an initial capability prediction model that has not yet completed parameter training. The capability prediction model to be trained has the same network structure as the capability prediction model, but the network parameters of the capability prediction model to be trained are in the initial state.

[0136] Iterative training can be a training process that uses historical feature data of multiple sample objects as input and real capability data of multiple sample objects as supervision labels, and repeatedly adjusts the network parameters of the capability prediction model to be trained through multiple iterations to minimize the prediction error.

[0137] Optionally, the terminal acquires historical feature data and real capability data of multiple sample objects. The historical feature data consists of fused feature data of multiple sample objects within a historical time period, while the real capability data consists of labeled real capability scores of multiple sample objects across various dimensions. The terminal uses the historical feature data as input and the real capability data as supervision labels to iteratively train the capability prediction model to be trained. In each iteration, the terminal inputs the historical feature data into the capability prediction model to obtain predicted capability data, calculates the loss value between the predicted capability data and the real capability data, and updates the network parameters of the capability prediction model to be trained through backpropagation based on the loss value. When the loss value converges or reaches the preset number of iterations, the terminal stops iterative training, thus obtaining the capability prediction model.

[0138] The technical solution provided in this embodiment obtains historical feature data and real capability data of multiple sample objects, and iteratively trains the capability prediction model to be trained based on the historical feature data and real capability data. This helps the capability prediction model to fully learn the mapping relationship between capability features and capability scores in historical data, thereby improving the prediction accuracy and generalization ability of the capability prediction model.

[0139] The following example illustrates the capability profile data generation method based on multi-source heterogeneous data fusion provided in this application. This example demonstrates the application of this method to a terminal.

[0140] This embodiment belongs to the interdisciplinary field of information technology and management science, specifically involving a method for intelligent fusion of multi-source heterogeneous talent data and construction of dynamic capability profiles for technological innovation scenarios. It is applicable to large enterprises, research institutions and other organizations for full life-cycle management of scientific and technological talents and precise allocation of innovation resources.

[0141] Currently, enterprises generally face the following technical bottlenecks in the management of scientific and technological talent:

[0142] (1) Serious data silos: Basic personnel information, project participation data, scientific research results, innovation and creativity, performance evaluation, etc. are scattered on independent platforms such as OA system, project life cycle management system (PLM), scientific research results database, human resources information system (HRIS), etc. There is a lack of a unified data collection and integration mechanism, which leads to talent evaluation relying on manual summarization, which is inefficient and prone to errors.

[0143] (2) The evaluation dimensions are too rigid and singular: Traditional talent evaluation focuses too much on explicit achievement indicators such as academic qualifications, professional titles, and papers, while neglecting implicit characteristics such as the depth of project practice, innovation potential, and cross-disciplinary collaboration ability, making it difficult to accurately identify high-potential innovative talents. The evaluation system mostly adopts fixed weights and cannot be dynamically adjusted according to the type of talent (such as young backbone, leading talents) or application scenario (project establishment, professional title evaluation), lacking flexibility and pertinence.

[0144] (3) High subjectivity and insufficient objectivity: Existing evaluations rely heavily on expert review and qualitative descriptions, lacking in-depth mining and quantitative analysis of unstructured data (such as project reports, technical documents, and creative proposals). The evaluation results are easily affected by subjective biases and cannot achieve dynamic tracking and continuous updating of talent capabilities.

[0145] (4) Poor real-time performance and weak decision support: Talent profiles are mostly static snapshots, which cannot synchronize the dynamic behavior data of talents in innovation activities in real time (such as new patents, project role changes, creative contributions), resulting in a lag in matching talents with projects and making it difficult to support the immediate talent needs of agile innovation mechanisms such as "challenge and take the lead".

[0146] (5) Fragmented technology implementation: Although some systems have introduced machine learning algorithms, they mostly remain at the level of single data source analysis and have not formed a closed-loop technology chain of data collection → cleaning and fusion → feature extraction → modeling and evaluation → visualization presentation → real-time synchronization. In addition, they lack standardized interfaces for deep integration with business systems, making it difficult to scale up applications.

[0147] This embodiment aims to construct a systematic, automated, and intelligent method for talent data fusion and profile building, with the following specific objectives:

[0148] (1) Break down data silos: Real-time collection, cleaning, transformation and fusion of multi-source heterogeneous data are achieved through automated ETL tools and named entity recognition (NER) technology to build a unified talent data foundation.

[0149] (2) Establish a scientific evaluation system: Design a four-dimensional progressive quantitative evaluation index system of "experience-professionalism-achievements-potential", combine the analytic hierarchy process (AHP) to dynamically allocate weights, and introduce neural networks to mine nonlinear relationships in data to improve the scientific nature and comprehensiveness of the evaluation.

[0150] (3) Achieve dynamic and accurate profiles: Based on integrated data and evaluation models, automatically generate a visual dynamic capability map, intuitively present the distribution of talent's professional and innovative capabilities, and synchronize with the business system in real time through standardized API interfaces to ensure that the profile is automatically updated along with the talent's growth trajectory.

[0151] (4) Supporting intelligent decision-making: providing real-time and accurate data support for scenarios such as talent matching, project team building, professional title evaluation, and innovation incentives, and realizing the transformation from "experience-based decision-making" to "data-driven decision-making".

[0152] Overall technical architecture:

[0153] This embodiment constructs a four-layer technical architecture: "data acquisition layer - fusion processing layer - evaluation modeling layer - profiling application layer", forming a closed loop across the entire chain.

[0154] Core module technical solution:

[0155] (1) Multi-source heterogeneous data acquisition and fusion module:

[0156] Data source access: Connect to OA systems, project lifecycle management systems, scientific research result databases, HR systems, financial systems, etc. through configurable ETL tools, supporting multiple access methods such as API interfaces, direct database connections, and file imports.

[0157] Data cleaning and transformation: Develop a dedicated data cleaning engine to automatically identify and correct noisy data, missing values, and erroneous data; use natural language processing technology to preprocess unstructured data such as project materials, certificates of honor, and technical documents to extract key text information.

[0158] Named Entity Recognition (NER) technology application: Using pre-trained language models (such as BERT-BiLSTM-CRF) to perform deep parsing of unstructured text, accurately extracting entities such as project names, technical roles, innovative achievements, and award levels, and establishing relationships between entities.

[0159] Data Fusion: Based on a unified talent master data model, multi-source data are linked and aligned according to "personnel ID + timestamp" to build a full-dimensional talent data warehouse covering basic information, education experience, work experience, project experience (including role, cycle, and results), scientific research results, innovation and creativity (proposals), performance appraisal, skills certificates, training records, etc.

[0160] (2) Construction of a comprehensive evaluation index system for scientific and technological innovation talents:

[0161] Four-dimensional progressive evaluation framework:

[0162] Experience dimension: assess the relevance of educational background to major, depth of research project experience, and breakthroughs in technological challenges.

[0163] Professional dimensions: assess competence for key technical positions, the value of rationalization proposals, the benchmark nature of operational standards, the radiating effect of talent cultivation, the authority of registration certificates, and the stability of performance appraisals over the past three years.

[0164] Results dimensions: Evaluation of the level of technology awards, academic influence of papers (citation frequency), value of intellectual property transformation, and industry recognition of standards and norms.

[0165] Potential dimensions: Innovation assessment: Innovation leadership (disruptive technological concepts), technological insight (industry trend prediction), systems thinking (cross-domain integration solutions), and resilience in overcoming difficulties (long-term R&D breakthroughs).

[0166] Dynamic weight allocation mechanism: The Analytic Hierarchy Process (AHP) is used to construct judgment matrices for different application scenarios. For example, in the selection of young talents, the weight of the potential dimension is increased to 40%, while the weight of the achievement dimension is reduced to 20%; in the evaluation of leading talents, the weight of the achievement dimension is increased to 50%, while the weight of the potential dimension is 15%. The weights are determined through expert scoring and consistency testing (CR<0.1) to ensure scientific rigor and rationality.

[0167] Quantifying openness indicators: For qualitative indicators of potential, behavioral anchor scales are designed to transform innovative behaviors into observable variables. For example, innovation leadership can be quantified through sub-indicators such as the number of forward-looking technology proposals submitted, the proportion of proposals incorporated into the company's strategy, and the adoption rate of cross-domain technology integration solutions.

[0168] (3) Construction of a capability assessment model based on neural networks:

[0169] Training dataset construction: A large-scale labeled dataset was constructed based on historical talent data and project outcome data. Each sample includes labels such as talent ID, four-dimensional indicator vector, historical project success rate, and innovative output.

[0170] Model architecture design: A hybrid neural network of multilayer perceptron (MLP) and gated recurrent unit (GRU) is adopted. MLP is used to process static features (education, professional title, etc.), and GRU is used to process temporal features (project experience, achievements accumulated over time). The two types of features are dynamically fused through the attention mechanism to output the comprehensive talent ability score and potential prediction value.

[0171] Nonlinear Relationship Mining: Residual connections and batch normalization are introduced to alleviate the vanishing gradient problem, deeply exploring the complex nonlinear relationships between "project roles - outputs - capability growth" to improve model prediction accuracy (target R²). 2 >0.85).

[0172] Model training and optimization: The Adam optimizer is used in conjunction with Dropout regularization to prevent overfitting; the early stopping strategy is used to automatically determine the optimal number of iterations; and the model's generalization ability is evaluated through K-fold cross-validation.

[0173] (4) Dynamic portrait generation and visualization:

[0174] Capability Radar Chart Generation: Based on the output of the assessment model, generate professional capability radar charts and innovation capability radar charts to intuitively show the strengths and weaknesses of talents in specific dimensions such as technology research and development, project management, and innovative thinking.

[0175] Dynamic update mechanism: An incremental update algorithm is designed so that when a new project record, achievement data, or creative contribution is added to the talent data warehouse, the evaluation model is automatically recalculated, and the profile parameters are updated. The update frequency is configurable (e.g., real-time, daily, weekly).

[0176] Interactive Visualization Interface: Employing front-end technologies such as ECharts, an interactive interface supporting drill-down, timeline navigation, and talent comparison is developed. Users can click on a dimension of the radar chart to view the supporting evidence chain (such as specific project results and creative content) for that dimension.

[0177] (5) Standardized API interface and real-time synchronization:

[0178] API Design: Following RESTful standards, it provides talent profile query interface, batch query interface, and profile update callback interface.

[0179] Real-time synchronization mechanism: Loosely coupled asynchronous communication is achieved using a message queue (such as Kafka). When new data is generated by the business system (such as project application or creative release), an event message is published to the queue; the profiling system consumes the message and triggers incremental updates, which are then pushed to the front end via WebSocket after completion, achieving real-time synchronization with a latency of up to seconds.

[0180] Access Control and Security: The interface uses OAuth 2.0 authentication and RBAC access control to ensure secure data access; the transport layer uses TLS encryption to prevent the leakage of sensitive information.

[0181] Deep integration with business scenarios:

[0182] In the "Challenge and Recruit" scenario: The system analyzes the project requirements, extracts the required skill tags and experience requirements, and generates a requirement vector; it calculates the cosine similarity between this vector and all profile vectors in the talent pool, and returns the Top-N matching candidates and the explanation of the matching degree (e.g., "This project requires heavy-load railway communication technology, and Zhang San led an LTE-R project in 2024, with a matching degree of 92%").

[0183] Project team building scenario: Based on the project role requirements (project leader, technical leader, core members), the profiling system is invoked to recommend candidates according to the "role-ability" matching model, and the complementarity and balance of the team's overall ability structure are evaluated.

[0184] In the professional title evaluation scenario: the system automatically captures the four-dimensional indicator data of talents within the evaluation period, generates a draft evaluation report, and reduces the workload of manual data entry and review.

[0185] Positive effects:

[0186] (1) Improved comprehensiveness and timeliness of data fusion: Through automated ETL and NER technologies, multi-source heterogeneous data can be fused in minutes, which improves efficiency by more than 90% compared with manual aggregation and achieves data integrity of more than 98%, thus solving the problem of data silos.

[0187] (2) Enhanced scientific rigor and objectivity of the assessment: The four-dimensional progressive indicator system and dynamic weighting mechanism cover both explicit and implicit abilities, and the assessment results are consistent with expert reviews by more than 85%; the neural network model explores deep correlations, and the accuracy of talent potential prediction reaches more than 80%, significantly reducing subjective bias.

[0188] (3) The profile is dynamic and has a significant visualization effect: the profile is updated in real time with innovation activities with a delay of less than 5 seconds; the interactive radar chart intuitively presents 20+ dimensions of capability characteristics, supports multi-dimensional comparative analysis, provides a visual basis for management decision-making, and improves decision-making efficiency by 60%.

[0189] (4) Strong accuracy and flexibility in business support: The standardized API interface enables seamless integration with 10+ business scenarios such as bidding and project application, the accuracy of talent recommendation and matching reaches over 90%, and the project team formation cycle is shortened from an average of 7 days to 2 days.

[0190] (5) Excellent scalability and adaptability: The system supports rapid access to new data sources (configuration ETL < 2 hours), the evaluation model can be continuously optimized through incremental learning, and the weight configuration can be flexibly adjusted to adapt to the talent management needs of different departments and different development stages.

[0191] For example:

[0192] Step 1: Data Acquisition and Fusion.

[0193] Input: The system automatically retrieved 3 technology projects (role: project leader) that employee Wang participated in in 2024 from the OA system, extracted a total project funding of 2.8 million yuan from the PLM system, obtained 2 authorized invention patents from the patent database, and obtained 5 golden ideas published by him from the creative platform (2 of which were adopted).

[0194] Processing: The NER engine parses the project description text and identifies key technical entities such as "heavy-haul railway" and "moving block"; the data fusion engine aggregates data by "Wang Gong ID + 2024" to generate structured records.

[0195] Output: Wang Gong's 2024 talent data package, which includes multi-dimensional information such as project experience, achievements, and creativity.

[0196] Step 2: Calculation of the indicator system.

[0197] Experience dimension: Wang has led two group-level projects as a project leader, scoring 85 points; his educational background is a doctorate, with a professional match rate of 95%, resulting in a weighted score of 90 points.

[0198] Professional dimension: Its suggestion of "braking optimization for heavy-haul trains" was adopted, saving 3 million yuan in operating costs annually, and scored 92 points based on the benefit weight.

[0199] Results dimension: Both invention patents have been commercialized and applied, with a score of 88 points in the results dimension.

[0200] Potential dimension: Three forward-looking technology ideas were proposed in 2024, one of which was incorporated into the company's plan, with a potential score of 85 points.

[0201] Dynamic weighting: Since the application scenario is "Annual Innovative Talent Selection", the weighting is "Achievements 30%, Potential 30%, Professionalism 25%, Experience 15%", and the overall score is 88×0.3+85×0.3+92×0.25+90×0.15=88.4 points.

[0202] Step 3: Image generation and visualization.

[0203] Radar Chart: A radar chart of Wang Gong's capabilities is generated, showing that "technological foresight" and "achievement transformation capability" are outstanding strengths, while "teamwork" is an area for improvement.

[0204] Trend analysis: Comparing data from 2022 to 2024, it shows that its "innovation leadership" has been increasing year by year, which is in line with the growth trajectory of high-potential talents.

[0205] Step 4: Business Applications.

[0206] In March 2025, the Ministry of Science and Technology released a list of candidates for “Key Technology Breakthroughs in Heavy-Haul Train Positioning Based on BeiDou.” After analyzing the requirements, the system calculated that Wang’s profile matched the requirements with a 91% match. The reasons for the recommendation were “experience in heavy-haul railway communication projects + 2 invention patents + forward-looking technical ideas”, ranking him second among the recommended candidates.

[0207] Alternative solutions:

[0208] Alternative Option 1 (Traditional Machine Learning Instead of Deep Learning): In the ability assessment model, Gradient Boosting Tree (GBDT) can be used to replace neural networks. GBDT has lower data requirements and faster training speed, but its ability to process time-series data is weaker than GRU. It is suitable for scenarios with small data scale (talent sample <1000) and low real-time requirements.

[0209] Alternative Solution 2 (Manual Rules Instead of AHP Weight Allocation): Weight allocation can be done using direct assignment based on expert experience, omitting the matrix operations and consistency checks of AHP. Its advantage is its simplicity, but it is highly subjective and lacks flexibility in weight adjustment. It is suitable for scenarios with stable organizational structures and unchanging evaluation criteria over a long period.

[0210] Alternative Solution 3 (Static Profiles Instead of Dynamic Updates): Profiles can be updated in batches on a monthly / quarterly basis, eliminating the need for real-time API synchronization and reducing system complexity. This solution has low technical implementation costs but cannot respond to immediate talent needs. It is suitable for traditional scientific research management scenarios with low-frequency innovation activities and long project cycles.

[0211] Alternative Solution 4 (Lightweight NER instead of heavyweight model): In the entity recognition stage, a lightweight method based on dictionary matching and regular expressions can be used to replace the BERT-BiLSTM-CRF model. The advantage is lower computational resource consumption, but the recognition accuracy decreases by approximately 15-20%. This approach is suitable for organizations with weak IT infrastructure and limited hardware resources.

[0212] Alternative Solution 5 (Dimensional Reduction Visualization as an Alternative to Radar Charts): Visualization can use dimensionality reduction algorithms (t-SNE / UMAP) to generate scatter plots of talent distribution, replacing radar charts. Scatter plots are better at identifying talent clusters and outliers, but they lose dimensional detail information and are suitable for strategic-level talent structure analysis rather than individual development guidance.

[0213] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0214] Based on the same inventive concept, this application also provides a capability profile data generation device based on multi-source heterogeneous data fusion for implementing the capability profile data generation method based on multi-source heterogeneous data fusion described above. The solution provided by this device is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more capability profile data generation device embodiments based on multi-source heterogeneous data fusion provided below can be found in the limitations of the capability profile data generation method based on multi-source heterogeneous data fusion described above, and will not be repeated here.

[0215] In one exemplary embodiment, such as Figure 4 As shown, a capability profile data generation device based on multi-source heterogeneous data fusion is provided. This capability profile data generation device 400 based on multi-source heterogeneous data fusion may include:

[0216] The data acquisition module 401 is used to acquire the raw data of the target object from multiple heterogeneous data sources; the raw data includes structured data and unstructured data.

[0217] The first input module 402 is used to input unstructured data into a pre-trained entity recognition model for entity recognition processing to obtain target entity data in the unstructured data.

[0218] The data fusion module 403 is used to perform data fusion processing on the target entity data and structured data to obtain the fused feature data of the target object;

[0219] The second input module 404 is used to input the fused feature data into the pre-trained capability prediction model for capability prediction processing, so as to obtain the capability prediction data of the target object in multiple dimensions.

[0220] The data generation module 405 is used to generate capability profile data of the target object based on capability prediction data.

[0221] In an exemplary embodiment, the second input module 404 is further configured to input the static feature data in the fused feature data into the static feature processing network in the capability prediction model for capability prediction processing to obtain the static feature vector of the target object; input the temporal feature data in the fused feature data into the temporal feature processing network in the capability prediction model for capability prediction processing to obtain the temporal feature vector of the target object; perform dynamic fusion processing on the static feature vector and the temporal feature vector through the attention mechanism layer in the capability prediction model to obtain the fused feature vector of the target object; and determine the capability prediction data based on the fused feature vector.

[0222] In an exemplary embodiment, the device 400 further includes: an information determination module, configured to acquire current application scenario information; determine weight information corresponding to each dimension based on the current application scenario information; and a second input module 404, configured to determine capability prediction data based on the fused feature vector and weight information.

[0223] In an exemplary embodiment, the multiple dimensions include experience dimension, professional dimension, achievement dimension and potential dimension; the data generation module 405 is further configured to generate capability distribution map data of the target object based on the capability prediction data of the target object under the experience dimension, professional dimension, achievement dimension and potential dimension; and determine capability profile data based on the capability distribution map data.

[0224] In an exemplary embodiment, the device 400 further includes: a result generation module, configured to acquire target demand information; perform feature extraction processing on the target demand information to obtain a demand feature vector; determine the matching degree between the target object and the target demand information based on the demand feature vector and capability profile data; and generate an object recommendation result corresponding to the target demand information based on the matching degree.

[0225] In an exemplary embodiment, the device 400 further includes: a model training module, configured to acquire historical feature data and real capability data of multiple sample objects; and to iteratively train the capability prediction model to be trained based on the historical feature data and real capability data to obtain the capability prediction model.

[0226] The modules in the aforementioned capability profile data generation device based on multi-source heterogeneous data fusion can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0227] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output interfaces, a communication interface, a display unit, and an input device. The processor, memory, and input / output interfaces are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interfaces are used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a capability profile data generation method based on multi-source heterogeneous data fusion. The display unit of the computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0228] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0229] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0230] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above-described method embodiments.

[0231] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0232] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0233] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0234] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for generating capability profile data based on multi-source heterogeneous data fusion, characterized in that, The method includes: The raw data of the target object is obtained from multiple heterogeneous data sources; the raw data includes structured data and unstructured data. The unstructured data is input into a pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data; The target entity data and the structured data are fused together to obtain the fused feature data of the target object. The fused feature data is input into a pre-trained capability prediction model for capability prediction processing to obtain capability prediction data of the target object in multiple dimensions. Based on the capability prediction data, capability profile data of the target object is generated.

2. The method according to claim 1, characterized in that, The step of inputting the fused feature data into a pre-trained capability prediction model for capability prediction processing, to obtain capability prediction data of the target object in multiple dimensions, includes: The static feature data in the fused feature data is input into the static feature processing network in the capability prediction model for capability prediction processing, thereby obtaining the static feature vector of the target object; The temporal feature data in the fused feature data is input into the temporal feature processing network in the capability prediction model for capability prediction processing, thereby obtaining the temporal feature vector of the target object; The static feature vector and the temporal feature vector are dynamically fused through the attention mechanism layer in the capability prediction model to obtain the fused feature vector of the target object. The capability prediction data is determined based on the fused feature vector.

3. The method according to claim 2, characterized in that, Before determining the capability prediction data based on the fused feature vector, the method further includes: Obtain information about the current application scenario; Based on the current application scenario information, determine the weight information corresponding to each dimension; The step of determining the capability prediction data based on the fused feature vector includes: The capability prediction data is determined based on the fused feature vector and the weight information.

4. The method according to claim 1, characterized in that, The multiple dimensions include experience, professional expertise, achievements, and potential. The step of generating capability profile data for the target object based on the capability prediction data includes: Based on the target object's predicted capabilities under the experience dimension, professional dimension, achievement dimension, and potential dimension, generate the target object's capability distribution map data; Based on the capability distribution map data, the capability profile data is determined.

5. The method according to claim 1, characterized in that, After generating the capability profile data of the target object based on the capability prediction data, the method further includes: Obtain target requirement information; The target demand information is subjected to feature extraction processing to obtain a demand feature vector; Based on the demand feature vector and the capability profile data, determine the degree of matching between the target object and the target demand information; Based on the matching degree, an object recommendation result corresponding to the target demand information is generated.

6. The method according to any one of claims 1 to 5, characterized in that, The capability prediction model is trained in the following way: Acquire historical feature data and real capability data of multiple sample objects; Based on the historical feature data and the actual capability data, the capability prediction model to be trained is iteratively trained to obtain the capability prediction model.

7. A capability profile data generation device based on multi-source heterogeneous data fusion, characterized in that, The device includes: The data acquisition module is used to acquire raw data of the target object from multiple heterogeneous data sources; the raw data includes structured data and unstructured data. The first input module is used to input the unstructured data into a pre-trained entity recognition model for entity recognition processing to obtain the target entity data in the unstructured data; The data fusion module is used to perform data fusion processing on the target entity data and the structured data to obtain the fused feature data of the target object; The second input module is used to input the fused feature data into a pre-trained capability prediction model for capability prediction processing, so as to obtain capability prediction data of the target object in multiple dimensions. The data generation module is used to generate capability profile data of the target object based on the capability prediction data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images