A method, system, and device for dynamically updating and recommending digital media talent competency maps based on incremental learning.

Abstract

This invention belongs to the field of generative natural language processing, and provides a method, system, and device for dynamically updating and recommending digital media talent literacy graphs based on incremental learning. It calculates the job knowledge coordinate sequence relative to the core of each job cluster, and the skill knowledge coordinate sequence of its associated skill entities relative to the core of each skill cluster, and generates a knowledge coordinate mapping matrix for job entities based on these two. When an incremental triple appears, it calculates a mapping difference index based on the knowledge coordinate mapping matrix of the corresponding job entity and the existing knowledge coordinate mapping matrix of job entities in the graph. The mapping difference index is compared with the mean and standard deviation of the mapping difference to determine whether to include the incremental triple in the knowledge graph. The incremental triple is written into the knowledge graph, the entity vector is updated, and a recommendation is output. By characterizing multiple principal axis gaps and adaptively adjusting scale and drift, it achieves second-level cold start and minute-level local integration.

Description

Technical Field

[0001] This invention belongs to the fields of natural language processing and generative artificial intelligence, specifically relating to a method, system, and device for dynamically updating and recommending digital media talent literacy graphs based on incremental learning. Background Technology

[0002] The system integrates platform data from recruitment websites and social media, including raw, unstructured data such as job descriptions, talent profiles, skills and resumes, and industry trend articles. These data streams are collected in real-time via web crawlers or APIs. The data processing and AI analysis layer includes modules for data acquisition and preprocessing, incremental learning models, concept drift detection mechanisms, graph embedding, and generative AI prediction. This layer cleans and analyzes the raw data, using a PyTorch-trained model for feature extraction and incremental learning analysis to identify new skill concepts and changes in relationships. The concept drift detection module continuously monitors changes in skill demand distribution over time, triggering model adjustments and graph updates. The generative AI module performs reasoning to complete missing nodes in the graph. The Neo4j graph database is used to store a knowledge graph semantically approximating the talent competency domain. The graph consists of nodes and relationships. Nodes represent elements related to skills, positions / occupations in the digital media industry, while relationships represent the dependencies between skills and positions, the correlations between skills, etc., and include weights and other attributes to quantify relationship strength. It provides a graphical knowledge graph visualization and management interface, including interactive network graph displays, query and retrieval interfaces, and analysis dashboards. The front-end retrieves updated graph data in real time and displays it to the user by calling back-end APIs or directly connecting to the graph database. Users can access the interface through desktop applications, web browsers, or mobile apps, enabling cross-platform operation.

[0003] With the rapid evolution of short videos, virtual production, and AIGC (AI-generated content), the skill requirements for digital media positions are updated monthly or even weekly. Companies typically release skill requirements for new tools and processes instantly through job postings or social networks; universities and training institutions, in turn, adjust their curricula accordingly. However, existing industry dictionaries and static skill ontologies, while comprehensive and standardized in terminology, suffer from update delays of several months, failing to reflect new skills like AIGC. Manually maintained job skill tables are typically compiled periodically by HR departments or third-party consulting firms, resulting in labor-intensive and costly updates. Embedded knowledge graphs, where some recruitment platforms extract job description text using NLP and embed it into graph databases using algorithms like TransE and Node2Vec, mostly require full retraining quarterly or semi-annually, lacking plug-and-play capabilities for cold-start positions.

[0004] Even with incremental extraction and knowledge graphs, existing dynamic update technologies still suffer from semantic vacuums in the incremental writing of new positions and skills. When a new position is first extracted, it lacks edges with any skill node in the graph, making it difficult to determine whether it can be written based on global vector distance. Recommendation systems may mistakenly delete or include entries, and the embedded model relies on neighborhood structures. For example, the knowledge graph-based human resource management system described in patent document CN117974080A suffers from cold-start adjacency issues and randomized initial vectors. Existing technologies suffer from high full-batch retraining costs. For instance, a knowledge graph recommendation method integrating GNN and ResNet described in patent document CN115114528A often requires rerunning the full-batch GNN after writing thousands of incremental entries, resulting in several hours of GPU time; this also affects online queries. Existing GNN / TransE training is parameterized by full node IDs, lacking local fine-tuning capabilities. Previously, mismatches between job positions and skill dimensions were not readily apparent. For example, the human resource data analysis method and system based on time-series knowledge graphs described in patent document CN120235597A only provides one-dimensional distances in its embeddings, failing to reveal gaps in job requirements across various skill axes. This makes it difficult for curriculum designers to address these gaps. Traditional similarity metrics lack multi-axis decomposition and cannot explain vector contributions. Existing incremental thresholds are subjective, relying heavily on fixed distance thresholds or manual review, and these thresholds become ineffective as the graph size and data distribution change. Furthermore, the lack of adaptive metrics makes them particularly unreliable for cross-domain and multi-skilled positions.

[0005] In the context of high-frequency iteration in digital media, the same job may involve both visual design and marketing planning. Traditional holistic vectors cannot reveal this duality, leading to misjudgments. Furthermore, quantifying the gaps in job demand and skill supply at the dimensional level is insufficient for educational institutions to determine which specific skill axis is lacking for a particular position. Current tools only provide a list without depth information on the gaps. Additionally, the reliability of incremental triplet adaptive thresholds becomes increasingly unreliable as the graph expands, lacking dynamic judgment methods based on graph statistics. How can plug-and-play, low-cost local fine-tuning, under GPU resource constraints, ensure that newly added nodes have usable representations within seconds and gradually integrate into the entire network without requiring database locking and retraining, given the limited resources available? Moreover, the "extract and write" approach lacks robustness checks, and outlier skills lack filtering mechanisms. Noise pollution, such as low-frequency or industry jargon, is easily extracted as skill nodes and written into the graph, causing link explosions. Summary of the Invention

[0006] The purpose of this invention is to propose a method, system, and device for dynamically updating and recommending digital media talent literacy maps based on incremental learning, so as to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create conditions.

[0007] To achieve the above objectives, according to one aspect of the present invention, a method for dynamically updating and recommending a digital media talent literacy map based on incremental learning is provided, the method comprising the following steps:

[0008] Step 1: Crawl and store job postings and skill descriptions from digital media platforms;

[0009] Step 2: Use a pre-trained language model to extract the text and obtain a set of triplets consisting of job entities, skill entities and the semantic relationship between them.

[0010] Step 3: Write the set of triples into the graph database to generate an initial talent literacy knowledge graph;

[0011] Step 4: Use a unified-dimensional knowledge graph embedding algorithm to vectorize each entity in the graph to obtain job entity vectors and skill entity vectors.

[0012] Step 5: Cluster the job entity vectors and skill entity vectors respectively to obtain the job cluster core set and the skill cluster core set;

[0013] Step 6: For each job entity, calculate its job knowledge coordinate sequence relative to each job cluster core, and its associated skill entity's skill knowledge coordinate sequence relative to each skill cluster core, and generate a knowledge coordinate mapping matrix for the job entity based on the two.

[0014] Step 7: When an incremental triple containing a newly added job entity or a newly added skill entity appears, calculate the mapping difference index based on the knowledge coordinate mapping matrix of the corresponding job entity and the knowledge coordinate mapping matrix of the existing job entities in the graph. Compare the mapping difference index with the threshold determined based on the mean and standard deviation of the mapping difference to determine whether to include the incremental triple in the knowledge graph.

[0015] Step 8: When the threshold condition is met, the incremental triple is written into the knowledge graph and the entity vector is updated; otherwise, the incremental triple is discarded.

[0016] Step 9: When the system user inputs the target job and skill description, the incremental alignment entity rate of the target job is obtained according to the calculation process of Steps 6 and 7, and the recommended jobs with the highest ranking are returned to the client.

[0017] Furthermore, the pre-trained language model in step two is a Chinese language model based on deep bidirectional encoding representation, and the job title, skill title, and the need relationship or correlation between the two are jointly extracted.

[0018] Furthermore, in step four, the knowledge graph embedding algorithm is selected as the transfer distance embedding algorithm that can handle symbolic triples, and the dimensions of each entity vector are consistent.

[0019] Furthermore, the clustering in step five adopts the density peak clustering algorithm, and uses the Euclidean distance between job entity vectors as a similarity measure.

[0020] Furthermore, the calculation methods for the job knowledge coordinate sequence and the skill knowledge coordinate sequence are as follows:

[0021] Calculate the reciprocal of the cross-entropy of the job entity vector relative to the cluster cores of each job entity, and use it as the job knowledge coordinates of the job entity relative to the cluster cores of each job entity. The sequence of the values ​​of the job knowledge coordinates of a job entity relative to the cluster cores of each job entity is the job knowledge coordinate mapping of the job entity.

[0022] The cross-entropy is calculated between the skill entity vectors of each skill entity associated with the job entity and each job skill cluster core. The reciprocal of the median value of the cross-entropy between the skill entity vectors of each skill entity associated with the job entity and a job skill cluster core is selected as the skill knowledge coordinate of the job entity relative to a job skill cluster core. The sequence of the values ​​of the skill knowledge coordinates of a job entity relative to each job skill cluster core is the skill knowledge coordinate mapping of the job entity.

[0023] Furthermore, the method for calculating the mapping discrepancy index is as follows:

[0024] Let the knowledge coordinate mapping matrix corresponding to the job entity in the incremental triple be called the incremental knowledge coordinate mapping matrix, and let the knowledge coordinate mapping matrix of the job entity that was originally in the talent literacy map that was aligned with it be called the stock knowledge coordinate mapping matrix.

[0025] The array of elements on the left diagonal of the incremental knowledge coordinate mapping matrix is ​​the left feature of the incremental knowledge coordinate mapping, and the array of elements on the right diagonal of the incremental knowledge coordinate mapping matrix is ​​the right feature of the incremental knowledge coordinate mapping.

[0026] The array formed by the elements on the left diagonal of the existing knowledge coordinate mapping matrix is ​​the left feature of the existing knowledge coordinate mapping, and the array formed by the elements on the right diagonal of the existing knowledge coordinate mapping matrix is ​​the right feature of the existing knowledge coordinate mapping.

[0027] The similarity between the left feature of the incremental knowledge coordinate mapping and the right feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate left-side feature, and the similarity between the right feature of the incremental knowledge coordinate mapping and the left feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate right-side feature. The mapping difference index is generated by combining the knowledge coordinate left-side feature and the knowledge coordinate right-side feature.

[0028] Furthermore, the mapping-matching difference index in step seven is generated by comparing the similarity between the left and right diagonal features of the incremental knowledge coordinate mapping matrix and the corresponding left and right diagonal features of the stock knowledge coordinate mapping matrix, and the mean and standard deviation of the mapping-matching difference are used for judgment.

[0029] This invention also provides a dynamic update and recommendation system for a digital media talent literacy graph based on incremental learning. The system includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the dynamic update and recommendation method for the digital media talent literacy graph based on incremental learning. This system can run on computing devices such as desktop computers, laptops, handheld computers, and cloud data centers. The runnable system may include, but is not limited to, processors, memory, and server clusters. The processor executes the computer program within the following system units:

[0030] The quantization unit is used to generate an initial talent literacy knowledge graph from the set of triples; and to vectorize each entity in the graph to obtain job entity vectors and skill entity vectors.

[0031] The mapping unit is used to perform clustering to obtain the core set of job clusters and the core set of skill clusters; calculate the job knowledge coordinate sequence relative to each job cluster core, and the skill knowledge coordinate sequence of its associated skill entity relative to each skill cluster core, and generate the knowledge coordinate mapping matrix of the job entity based on the two.

[0032] The judgment unit is used to calculate the mapping difference index based on the knowledge coordinate mapping matrix of the corresponding job entity and the knowledge coordinate mapping matrix of the existing job entities in the graph when an incremental triplet appears, and to compare the mapping difference index with the mean and standard deviation of the mapping difference to determine whether to include the incremental triplet in the knowledge graph.

[0033] The recommendation unit is used to write the incremental triples into the knowledge graph and update the entity vectors when the judgment conditions are met, and then output recommendations.

[0034] Correspondingly, the present invention also provides an electronic device, a readable storage medium, and a computer program product:

[0035] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the incremental learning-based digital media talent literacy map dynamic update recommendation method and the method for each step therein.

[0036] A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the incremental learning-based digital media talent literacy map dynamic update recommendation method and the methods for each step therein.

[0037] A computer program product includes a computer program that, when executed by a processor, implements the incremental learning-based method for dynamically updating and recommending digital media talent literacy maps, as well as the methods for each step therein.

[0038] The beneficial effects of this invention are as follows: This invention provides a method, system, and device for dynamically updating and recommending digital media talent literacy graphs based on incremental learning. It calculates the job knowledge coordinate sequence relative to the core of each job cluster, and the skill knowledge coordinate sequence of its associated skill entities relative to the core of each skill cluster. Based on these two, a knowledge coordinate mapping matrix for the job entities is generated. When an incremental triple appears, a mapping difference index is calculated based on the knowledge coordinate mapping matrix of the corresponding job entity and the existing knowledge coordinate mapping matrix of job entities in the graph. This mapping difference index is compared with the mean and standard deviation of the mapping difference to determine whether the incremental triple should be included in the knowledge graph. The incremental triple is written into the knowledge graph, the entity vector is updated, and a recommendation is output. By characterizing multiple principal axis gaps and adaptively adjusting scale and drift, it achieves second-level cold start and minute-level local integration, ultimately balancing real-time performance, accuracy, and interpretability without requiring full retraining. Attached Figure Description

[0039] The above and other features of the present invention will become more apparent from the detailed description of the embodiments shown in conjunction with the accompanying drawings. In the accompanying drawings, the same reference numerals denote the same or similar elements. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort. In the drawings:

[0040] Figure 1 The diagram shows a flowchart of a dynamic update and recommendation method for a digital media talent competency graph based on incremental learning.

[0041] Figure 2The diagram shows the system architecture of a recommendation system for dynamically updating digital media talent literacy graphs based on incremental learning. Detailed Implementation

[0042] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0043] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0044] like Figure 1 The diagram shown is a flowchart of the dynamic update and recommendation method for the digital media talent literacy map based on incremental learning according to the present invention. The following is a combination of... Figure 1 This paper describes a method, system, and device for dynamically updating and recommending digital media talent literacy maps based on incremental learning, according to embodiments of the present invention.

[0045] This invention proposes a dynamic update and recommendation method for a digital media talent literacy graph based on incremental learning. The method specifically includes the following steps:

[0046] The set of triples is used to generate an initial talent competency knowledge graph; each entity in the graph is vectorized to obtain job entity vectors and skill entity vectors.

[0047] Clustering is performed separately to obtain the core set of job clusters and the core set of skill clusters; the job knowledge coordinate sequence relative to each job cluster core and the skill knowledge coordinate sequence of its associated skill entity relative to each skill cluster core are calculated, and a knowledge coordinate mapping matrix of job entities is generated based on the two.

[0048] When an incremental triplet appears, the mapping difference index is calculated based on the knowledge coordinate mapping matrix of the corresponding job entity and the knowledge coordinate mapping matrix of the existing job entities in the graph. The mapping difference index is compared with the mean and standard deviation of the mapping difference to determine whether the incremental triplet should be included in the knowledge graph.

[0049] When the judgment condition is met, the incremental triple is written into the knowledge graph and the entity vector is updated for recommendation output.

[0050] Furthermore, a pre-trained language model is used in the extraction of the text. The pre-trained language model is a language model based on deep bidirectional encoding representation, and the job title, skill title, and the need or correlation relationship between the two are jointly extracted.

[0051] Furthermore, a knowledge graph embedding algorithm is used to vectorize the entities in the graph. The knowledge graph embedding algorithm uses the transfer distance embedding algorithm that processes symbol triples, and the dimensions of each entity vector are consistent.

[0052] Furthermore, in the clustering of job entity vectors and skill entity vectors, the density peak clustering algorithm is used, and the Euclidean distance between job entity vectors is used as the similarity measure.

[0053] Furthermore, the method for generating the knowledge coordinate mapping matrix of the job entity is as follows:

[0054] In the talent competency map, for each job entity, the job knowledge coordinate mapping and the skill knowledge coordinate mapping of each job entity are obtained respectively.

[0055] The vector formed by comparing the values ​​of each dimension in the job knowledge coordinate mapping of a job entity with the values ​​of each dimension in the skill knowledge coordinate mapping of the same job entity is one column of a matrix, and the vector formed by comparing the values ​​of each dimension in the job knowledge coordinate mapping with the values ​​of each dimension in the skill knowledge coordinate mapping of the same job entity is one column of a matrix. The matrix formed by these ratios is the knowledge coordinate mapping matrix of the job entity.

[0056] Furthermore, the calculation methods for the job knowledge coordinate sequence and the skill knowledge coordinate sequence are as follows:

[0057] Calculate the reciprocal of the cross-entropy of the job entity vector relative to the cluster cores of each job entity, and use it as the job knowledge coordinates of the job entity relative to the cluster cores of each job entity. The sequence of the values ​​of the job knowledge coordinates of a job entity relative to the cluster cores of each job entity is the job knowledge coordinate mapping of the job entity.

[0058] The cross-entropy is calculated between the skill entity vectors of each skill entity associated with the job entity and each job skill cluster core. The reciprocal of the median value of the cross-entropy between the skill entity vectors of each skill entity associated with the job entity and a job skill cluster core is selected as the skill knowledge coordinate of the job entity relative to the job skill cluster core. The sequence of the values ​​of the skill knowledge coordinates of a job entity relative to each job skill cluster core is the skill knowledge coordinate mapping of the job entity.

[0059] Furthermore, the method for calculating the mapping discrepancy index is as follows:

[0060] Let the knowledge coordinate mapping matrix corresponding to the job entity in the incremental triple be called the incremental knowledge coordinate mapping matrix, and let the knowledge coordinate mapping matrix of the job entity that was originally in the talent literacy map that was aligned with it be called the stock knowledge coordinate mapping matrix.

[0061] The array of elements on the left diagonal of the incremental knowledge coordinate mapping matrix is ​​the left feature of the incremental knowledge coordinate mapping, and the array of elements on the right diagonal of the incremental knowledge coordinate mapping matrix is ​​the right feature of the incremental knowledge coordinate mapping.

[0062] The array formed by the elements on the left diagonal of the existing knowledge coordinate mapping matrix is ​​the left feature of the existing knowledge coordinate mapping, and the array formed by the elements on the right diagonal of the existing knowledge coordinate mapping matrix is ​​the right feature of the existing knowledge coordinate mapping.

[0063] The similarity between the left feature of the incremental knowledge coordinate mapping and the right feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate left-side feature, and the similarity between the right feature of the incremental knowledge coordinate mapping and the left feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate right-side feature. The mapping difference index is generated by combining the knowledge coordinate left-side feature and the knowledge coordinate right-side feature.

[0064] In some embodiments, multiple different job postings are crawled from digital media. These job postings are structured or semi-structured text data. A pre-trained language model is used to extract entity relationships from these multiple job postings. Triples are extracted, consisting of text entities representing job types (job entities), text entities representing required skills (skill entities), and semantic relationships between the two types of text entities. A knowledge graph composed of multiple extracted triples is used as a talent competency graph. The text entities are all text data in string format. The text entities may differ slightly from each other but cannot be completely identical. A text entity representing a job type can be connected to multiple different text entities representing required skills through multiple different semantic relationships, and a text entity representing required skills can be connected to multiple different text entities representing job types through multiple different semantic relationships.

[0065] From the talent competency map, the set of text entities representing job types is obtained as the job entity set, and the set of text entities representing the skills required for the job is obtained as the skill entity set.

[0066] Using knowledge graph vectorization embedding algorithms such as TransE, the dimensions of each embedding vector are kept uniform. The triples in the talent competency graph are vectorized and embedded to obtain the embedding vectors of each text entity. The embedding vectors corresponding to the text entities in the job entity set are job entity vectors, and the embedding vectors corresponding to the text entities in the skill entity set are skill entity vectors.

[0067] Clustering the job entity vectors yields multiple different job entity cluster cores, and clustering the skill entity vectors yields multiple different job skill cluster cores.

[0068] In some embodiments, in the talent competency graph, for each job entity, skill entities that have a semantic relationship with that job entity are obtained;

[0069] Then, the reciprocal of the cross-entropy of the job entity vector relative to the cluster cores of each job entity is calculated, and these reciprocals are used as the job knowledge coordinates of the job entity relative to each job entity cluster core. The sequence of job knowledge coordinate values ​​of a job entity relative to each job entity cluster core is the job knowledge coordinate mapping of the job entity. For example, if there are n different job entity cluster cores in the talent literacy map, then a job entity has n job knowledge coordinate values ​​relative to each job entity cluster core, and the job knowledge coordinate mapping of the job entity has n dimensions.

[0070] Next, the cross-entropy is calculated between the skill entity vectors of each skill entity associated with the job entity and each job skill cluster core. The reciprocal of the median value of the cross-entropy between the skill entity vectors of each skill entity associated with the job entity and a job skill cluster core is selected as the skill knowledge coordinate of the job entity relative to the job skill cluster core. The sequence of skill knowledge coordinate values ​​of a job entity relative to each job skill cluster core is the skill knowledge coordinate mapping of the job entity. For example, if there are m job skill cluster cores in the talent literacy map, then the skill entity vectors of each skill entity associated with a job entity have m skill knowledge coordinate values ​​relative to each job skill cluster core, that is, the skill knowledge coordinate mapping of a job entity has m dimensions.

[0071] Compared to existing technologies that only use global Euclidean / cosine distance after graph embedding, which struggles to explain the multidimensional similarity between job positions and clusters, this invention addresses the issue of job knowledge coordinates. When a job crosses domains, such as design + marketing, the contribution of embedding distance to the "design" or "marketing" dimension cannot be separated, leading to distorted job profiles. This invention decomposes the one-dimensional distance into n-dimensional coordinates, quantifying the job relative to n core job clusters, thus pinpointing the job's gaps and strengths along each competency axis. Job knowledge coordinates resolve cross-domain job profile distortion and support dimensional support for teaching or training suggestions. The use of the inverse of cross-entropy naturally measures the overlap with the distribution of a cluster core, ensuring that each dimension of the sequence closely reflects the semantic fit between the job and that axis. After serialization, vector difference operations can be directly performed to identify dimensional gaps without needing to revert to textual interpretation.

[0072] Most existing systems only compile a list of job-related skills, failing to measure the saturation of these skills relative to the industry's skill axis. Furthermore, the length of job-related skills in existing systems makes it difficult for recommendation systems to distinguish between core and supplementary skills. The skill knowledge coordinates described in this invention, however, map skills to the same k skill clustering cores, resulting in an m-dimensional sequence. The numerical values ​​directly reflect the coverage depth of the required skills across each axis. Combined with the job sequence, job-skill alignment can be calculated, accurately predicting missing skills. The reciprocal of the median cross-entropy between the skill vector and the clustering cores automatically suppresses outliers; a maximum cross-entropy causes the reciprocal to approach zero, highlighting core skills. The median's superior noise immunity compared to the mean prevents the sequence from being skewed by a few abnormal skills.

[0073] In some embodiments, in the talent competency map, for each job entity, the job knowledge coordinate mapping of each job entity and the skill knowledge coordinate mapping of each job entity are obtained respectively;

[0074] The vector formed by comparing the values ​​of each dimension in the job knowledge coordinate mapping of a job entity with the values ​​of each dimension in the skill knowledge coordinate mapping of the same job entity is one column of a matrix, and the vector formed by comparing the values ​​of each dimension in the job knowledge coordinate mapping with the values ​​of each dimension in the skill knowledge coordinate mapping of the same job entity is one column of a matrix. The matrix formed by these ratios is the knowledge coordinate mapping matrix of the job entity.

[0075] In Example A, when new job posting information is crawled from digital media, the pre-trained language model is used to extract entity relations from the new job posting information, extracting triples composed of job entity, skill entity, and semantic relations between the two types of text entities. When a triple that has not appeared in the talent literacy graph appears, or when a triple contains a text entity that has not appeared in the talent literacy graph, it is called an incremental triple. The knowledge graph vectorization embedding algorithm is also used to vectorize and embed the incremental triples to obtain the embedding vector of each text entity.

[0076] Obtain the job entity and skill entity in the incremental triples, and then obtain the corresponding job entity vector and skill entity vector in the incremental triples. Similarly, calculate the knowledge coordinate mapping matrix corresponding to the job entity in the incremental triples based on the job entity cluster core and the job skill cluster core.

[0077] Traditional methods can only compare the overall similarity of job vectors, failing to reveal whether the job dimension and skill dimension are consistent, and lacking tools to quantify the alignment error between job demand and existing skill dimensions. The knowledge coordinate mapping matrix described in this invention is an n×m matrix that compares the job coordinate sequence and the skill coordinate sequence pairwise. The left and right diagonals capture the orthogonality and anti-orthogonality of demand and supply. After matrixing, it clearly identifies which skill dimensions have insufficient or redundant supply. A ratio >1 indicates insufficient skill supply, and <1 indicates excess. The left diagonal represents the matching degree along the same axis, and the right diagonal represents complementarity or mismatch. When using the knowledge coordinate mapping matrix for incremental alignment, only the diagonal vectors need to be extracted to significantly compress the feature dimensions, avoiding recalculating the full-dimensional distance.

[0078] The specific method for calculating the mapping difference value is as follows:

[0079] Align the knowledge coordinate mapping matrices corresponding to the job entities in the incremental triples with the existing knowledge coordinate mapping matrices of the job entities in the talent competency graph.

[0080] Then, the knowledge coordinate mapping matrix corresponding to the job entity in the incremental triplet is called the incremental knowledge coordinate mapping matrix, and the knowledge coordinate mapping matrix of the job entity that was originally in the talent literacy map that it is aligned with is called the stock knowledge coordinate mapping matrix.

[0081] The array of elements on the left diagonal of the incremental knowledge coordinate mapping matrix is ​​the left feature of the incremental knowledge coordinate mapping, and the array of elements on the right diagonal of the incremental knowledge coordinate mapping matrix is ​​the right feature of the incremental knowledge coordinate mapping.

[0082] The array formed by the elements on the left diagonal of the existing knowledge coordinate mapping matrix is ​​the left feature of the existing knowledge coordinate mapping, and the array formed by the elements on the right diagonal of the existing knowledge coordinate mapping matrix is ​​the right feature of the existing knowledge coordinate mapping.

[0083] The similarity between the left feature of the incremental knowledge coordinate mapping and the right feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate left-side feature, and the similarity between the right feature of the incremental knowledge coordinate mapping and the left feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate right-side feature. The knowledge coordinate left-side feature and the knowledge coordinate right-side feature are combined, for example by multiplication, to generate the numerical value of the mapping difference index of the job entity in the incremental triplet relative to the original job entity.

[0084] Furthermore, the mapping difference between each job entity in the incremental triplet and each existing job entity is calculated, and the arithmetic mean of the mapping difference values ​​of each job entity in the incremental triplet and each existing job entity is calculated as the mapping difference mean. The standard deviation of the mapping difference values ​​of each job entity in the incremental triplet and each existing job entity is calculated as the mapping difference standard deviation.

[0085] The mapping difference values ​​of each existing job entity in the talent competency map are compared to see if they are greater than the mean mapping difference plus the standard deviation of the mapping difference. If so, the existing job entity is marked as the incremental aligned job entity in the incremental triplet. The proportion of incremental aligned job entities to existing job entities in the talent competency map is calculated as the incremental aligned entity rate. The ratio of the standard deviation of the mapping difference to the mean mapping difference is calculated as the mapping difference entity rate. It is determined whether the incremental aligned entity rate is greater than or equal to the mapping difference entity rate. If so, the incremental triplet is added to the talent competency map; otherwise, the incremental triplet is not added to the talent competency map, thereby updating the talent competency map.

[0086] Previous technologies often used embedding distance thresholds to determine whether incremental nodes should be written into the graph. However, these thresholds are difficult to adjust, drift with the corpus, and ignore complementary information from the left and right diagonals, leading to false rejections or acceptances of cold-start positions. The mapping-matching difference index described in this invention combines the similarity of the left features of the incremental matrix and the right features of the existing matrix with the similarity of the right features of the incremental matrix and the left features of the existing matrix to obtain a single mapping-matching difference value. This simultaneously measures both matching and mismatch, making it more robust to cross-domain and composite positions. An adaptive threshold is formed by combining the mean and standard deviation, automatically calibrating as the graph size changes. Specifically, left features compared to right features can detect whether the job demand dimension is covered by the skill supply dimension; right features compared to left features can detect whether skills deviate from the main demand; the product of the two naturally amplifies cases where both are good or bad, suppressing unidirectional noise. The subsequent addition and subtraction of the mean and threshold is equivalent to normalization, automatically offsetting differences in sample size across different batches. "Mapping" represents mapping multidimensional demand and supply into a matrix, "matching" represents calculating coaxial fit, and "difference" represents the final comprehensive difference index.

[0087] In Embodiment B, for example, job seekers and other system users can input string information in the client of the system described in this invention, including their expected job type and skill description, through methods such as GUI, App typing, or language input converted to text. This string information input on the client is used as the newly added job recruitment information in Embodiment B. The incremental triples are obtained in the same way as in Embodiment A, denoted as incremental triple B. The mapping difference values ​​of the job entities in incremental triple B relative to each existing job entity in the talent literacy map are calculated. The mean and standard deviation of the mapping difference of the job entities in incremental triple B relative to each existing job entity are also calculated. By marking the incremental aligned job entities of the job entities in incremental triple B in the talent literacy map, the incremental aligned entity rate corresponding to incremental triple B is statistically calculated.

[0088] The incremental aligned job entities in the incremental triple B of the talent competency map are sorted from largest to smallest. Based on the incremental aligned entity rate corresponding to the incremental triple B, the incremental aligned job entities ranked first among the incremental aligned job entities in the incremental triple B after sorting are obtained as recommended incremental aligned job entities and recommended to the client of the system described in this invention.

[0089] For example, if the incremental alignment entity rate corresponding to the incremental triple B is approximately 0.03, and the incremental alignment job entities in the incremental triple B are sorted from largest to smallest, there are a total of 100 incremental alignment job entities, then the information of the top 3% of these 100 incremental alignment job entities is recommended to be sent to the system's client.

[0090] The incremental learning-based dynamic update and recommendation system for digital media talent literacy graphs runs on any computing device, such as a desktop computer, laptop computer, handheld computer, or cloud data center. The computing device includes a processor, a memory, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps in the incremental learning-based dynamic update and recommendation method for digital media talent literacy graphs. The runnable system may include, but is not limited to, processors, memory, and server clusters.

[0091] The embodiments of the present invention provide a dynamic update recommendation system for digital media talent literacy graphs based on incremental learning, such as... Figure 2 As shown, the incremental learning-based dynamic update recommendation system for digital media talent literacy graphs in this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps described in the above embodiment of the incremental learning-based dynamic update recommendation method for digital media talent literacy graphs. The processor executes the computer program within the following system units:

[0092] The quantization unit is used to generate an initial talent literacy knowledge graph from the set of triples; and to vectorize each entity in the graph to obtain job entity vectors and skill entity vectors.

[0093] The mapping unit is used to perform clustering to obtain the core set of job clusters and the core set of skill clusters; calculate the job knowledge coordinate sequence relative to each job cluster core, and the skill knowledge coordinate sequence of its associated skill entity relative to each skill cluster core, and generate the knowledge coordinate mapping matrix of the job entity based on the two.

[0094] The judgment unit is used to calculate the mapping difference index based on the knowledge coordinate mapping matrix of the corresponding job entity and the knowledge coordinate mapping matrix of the existing job entities in the graph when an incremental triplet appears, and to compare the mapping difference index with the mean and standard deviation of the mapping difference to determine whether to include the incremental triplet in the knowledge graph.

[0095] The recommendation unit is used to write the incremental triples into the knowledge graph and update the entity vectors when the judgment conditions are met, and then output recommendations.

[0096] In order to better unify the linear relationship and probabilistic connection between physical quantities with different units of measurement, dimensionless processing can be performed on different physical quantities.

[0097] Preferably, all undefined variables in this invention, if not explicitly defined, can be manually set thresholds.

[0098] The incremental learning-based dynamic update and recommendation system for digital media talent literacy graphs can run on computing devices such as desktop computers, laptops, handheld computers, and cloud data centers. This system includes, but is not limited to, processors and memory. Those skilled in the art will understand that the examples described are merely illustrations of the incremental learning-based dynamic update and recommendation method, system, and device for digital media talent literacy graphs, and do not constitute a limitation on the method, system, and device. It may include more or fewer components, or combine certain components, or different components. For example, the incremental learning-based dynamic update and recommendation system for digital media talent literacy graphs may also include input / output devices, network access devices, buses, etc.

[0099] The present invention also provides an electronic device, a readable storage medium, and a computer program product:

[0100] An electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the incremental learning-based digital media talent literacy map dynamic update recommendation method and the method for each step therein.

[0101] A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the incremental learning-based digital media talent literacy map dynamic update recommendation method and the methods for each step therein.

[0102] A computer program product includes a computer program that, when executed by a processor, implements the incremental learning-based method for dynamically updating and recommending digital media talent literacy maps, as well as the methods for each step therein.

[0103] The term "electronic device" is intended to refer to various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also refer to various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0104] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0105] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0106] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0107] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0108] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0109] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

[0110] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete component gate circuits, transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. This processor is the control center of the incremental learning-based dynamic update and recommendation system for digital media talent literacy maps, connecting various sub-regions of the system via various interfaces and lines.

[0111] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the incremental learning-based digital media talent literacy map dynamic update recommendation method, system, and device. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function, such as sound playback function, image playback function, etc.; the data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory and non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0112] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0113] This invention provides a method, system, and device for dynamically updating and recommending a digital media talent literacy graph based on incremental learning. The method involves crawling and storing job posting information and skill description text from a digital media platform; extracting the text using a pre-trained language model to obtain a set of triples consisting of job entities, skill entities, and the semantic relationships between them; writing the triple set into a graph database to generate an initial talent literacy knowledge graph; using a unified-dimensional knowledge graph embedding algorithm to vectorize each entity in the graph to obtain job entity vectors and skill entity vectors; clustering the job entity vectors and skill entity vectors to obtain a core set of job clusters and a core set of skill clusters; and for each job entity, calculating its job knowledge coordinate sequence relative to each job cluster core, as well as its associated skill coordinates. The system generates a knowledge coordinate mapping matrix for job entities based on the skill knowledge coordinate sequence of each skill cluster core. When an incremental triple containing a newly added job entity or a newly added skill entity appears, the system calculates the mapping difference index based on the knowledge coordinate mapping matrix of the corresponding job entity and the knowledge coordinate mapping matrix of existing job entities in the knowledge graph. The mapping difference index is compared with a threshold determined based on the mean and standard deviation of the mapping difference to determine whether to include the incremental triple in the knowledge graph. If the threshold condition is met, the incremental triple is written into the knowledge graph and the entity vector is updated; otherwise, the incremental triple is discarded. When the system user inputs a target job and skill description, the system obtains the incremental alignment entity rate of the target job according to the calculation process in steps six and seven, and returns the top-ranked recommended jobs to the client.

[0114] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A dynamic update and recommendation method for digital media talent literacy graph based on incremental learning, characterized in that, The method includes: The process involves crawling text containing job postings and skill descriptions; extracting the text to obtain a set of triples consisting of job entities, skill entities, and the semantic relationships between them; generating an initial knowledge graph from the set of triples; and vectorizing each entity in the knowledge graph to obtain job entity vectors and skill entity vectors. Cluster the job entity vectors and skill entity vectors separately to obtain the set of job cluster cores and the set of skill cluster cores; calculate the job knowledge coordinate sequence of each job cluster core and the skill knowledge coordinate sequence of each skill cluster core, and generate the knowledge coordinate mapping matrix corresponding to the job entity based on the job knowledge coordinate sequence and the skill knowledge coordinate sequence. When an incremental triplet appears, the mapping difference is calculated based on the knowledge coordinate mapping matrix of the job entity corresponding to the incremental triplet and the knowledge coordinate mapping matrix of the existing job entities in the knowledge graph. Whether the mapping difference is greater than the sum of the mean and standard deviation of the mapping difference is used to determine whether the incremental triplet should be included in the knowledge graph. The mapping difference is generated by comparing the features of the left and right diagonals of the incremental knowledge coordinate mapping matrix with the features of the corresponding existing knowledge coordinate mapping matrix. When the judgment condition is met, the incremental triple is written into the knowledge graph and updated, and a recommendation is output. The method for generating the knowledge coordinate mapping matrix of job entities is as follows: In the talent competency map, for each job entity, the job knowledge coordinate mapping and the skill knowledge coordinate mapping of each job entity are obtained respectively. The vector formed by comparing the values ​​of each dimension in the job knowledge coordinate mapping of a job entity with the values ​​of each dimension in the skill knowledge coordinate mapping of the same job entity is one column of a matrix. The vector formed by comparing the values ​​of each dimension in the job knowledge coordinate mapping with the values ​​of each dimension in the skill knowledge coordinate mapping of the same job entity is one column of a matrix. The matrix formed by these ratios is the knowledge coordinate mapping matrix of the job entity. The calculation methods for the job knowledge coordinate sequence and the skill knowledge coordinate sequence are as follows: Calculate the reciprocal of the cross-entropy of the job entity vector relative to the cluster cores of each job entity, and use it as the job knowledge coordinates of the job entity relative to the cluster cores of each job entity. The sequence of the values ​​of the job knowledge coordinates of a job entity relative to the cluster cores of each job entity is the job knowledge coordinate mapping of the job entity. The cross-entropy is calculated between the skill entity vectors of each skill entity associated with the job entity and each job skill cluster core. The reciprocal of the median value of the cross-entropy between the skill entity vectors of each skill entity associated with the job entity and the job skill cluster core is selected as the skill knowledge coordinate of the job entity relative to a job skill cluster core. The sequence of the values ​​of the skill knowledge coordinates of the job entity relative to each job skill cluster core is the skill knowledge coordinate mapping of the job entity.

2. The method for dynamically updating and recommending digital media talent literacy maps based on incremental learning according to claim 1, characterized in that, A pre-trained language model is used to extract the text. The pre-trained language model is a language model based on deep bidirectional encoding representation, and the job title, skill title and the need relationship or correlation between the two are jointly extracted.

3. The method for dynamically updating and recommending digital media talent literacy maps based on incremental learning according to claim 1, characterized in that, The knowledge graph embedding algorithm is used to vectorize the entities in the graph. The knowledge graph embedding algorithm is the migration distance embedding algorithm that processes symbol triples, and the dimensions of each entity vector are consistent.

4. The method for dynamically updating and recommending digital media talent literacy maps based on incremental learning according to claim 3, characterized in that, In the clustering of job entity vectors and skill entity vectors, density peak clustering algorithm is used, and the Euclidean distance between job entity vectors is used as the similarity measure.

5. The method for dynamically updating and recommending digital media talent literacy maps based on incremental learning according to claim 1, characterized in that, in, The method for calculating the mapping difference is as follows: The knowledge coordinate mapping matrix corresponding to the job entity in the incremental triple is the incremental knowledge coordinate mapping matrix, while the knowledge coordinate mapping matrix of the existing job entity in the knowledge graph is the stock knowledge coordinate mapping matrix. The array of elements on the left diagonal of the incremental knowledge coordinate mapping matrix is ​​the left feature of the incremental knowledge coordinate mapping, and the array of elements on the right diagonal of the incremental knowledge coordinate mapping matrix is ​​the right feature of the incremental knowledge coordinate mapping. The array formed by the elements on the left diagonal of the existing knowledge coordinate mapping matrix is ​​the left feature of the existing knowledge coordinate mapping, and the array formed by the elements on the right diagonal of the existing knowledge coordinate mapping matrix is ​​the right feature of the existing knowledge coordinate mapping. The similarity between the left feature of the incremental knowledge coordinate mapping and the right feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate left-side feature, and the similarity between the right feature of the incremental knowledge coordinate mapping and the left feature of the existing knowledge coordinate mapping is calculated as the knowledge coordinate right-side feature. The mapping difference is generated by combining the knowledge coordinate left-side feature and the knowledge coordinate right-side feature.

6. A dynamic update and recommendation system for digital media talent literacy graph based on incremental learning, characterized in that: The incremental learning-based dynamic update and recommendation system for digital media talent literacy graphs runs on any computing device, such as a desktop computer, a laptop computer, or a cloud data center. The computing device includes a processor, a memory, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the incremental learning-based dynamic update and recommendation method for digital media talent literacy graphs as described in any one of claims 1 to 5.

7. An electronic device, characterized in that, include: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 5.

Images