Object matching method based on attribute alignment and feature fusion

Abstract

The embodiment of the application provides an object matching method based on attribute alignment and feature fusion, which comprises: obtaining attribute description text data of a first target object, and performing feature extraction thereon to obtain a plurality of first attribute description feature word groups; matching the plurality of first attribute description feature word groups with a set attribute classification table to cluster the first attribute description feature word groups, and obtaining a plurality of first dimension attribute feature sequences for describing the first target object; performing vectorization on each first dimension attribute feature sequence to obtain a corresponding first attribute feature vector to form a feature portrait of the first target object; obtaining attribute description text data of a second target object, and performing feature extraction thereon to obtain a plurality of second attribute description feature word groups to form a feature portrait of the second target object, so as to determine whether the first target object and the second target object match.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an object matching method based on attribute alignment and feature fusion. Background Technology

[0002] With the development of the internet, various business models based on online platforms are constantly emerging. The company's influencer database, as an important resource, plays a crucial role in many business scenarios such as marketing promotion and content dissemination. To achieve efficient utilization of influencer resources and accurately recommend influencers to suitable partners (such as brands and event organizers), recommendation algorithms have become the core technological support.

[0003] In the field of existing recommendation algorithms, collaborative filtering is widely used, mainly divided into two common modes: user-based recommendation and item-based recommendation. For example, in user-based recommendation, if user A likes item P1 and user B shows similarity to user A in some aspects, item P1 will be recommended to user B. In item-based recommendation, if user A likes item P1 and item P2 has similar features to item P1, then item P2 will be recommended to user A.

[0004] However, regardless of the collaborative filtering mode, traditional similarity algorithms such as the dot product function, the law of cosines, or the Pearson coefficient are commonly used in the crucial step of calculating object similarity. These algorithms have significant limitations in application, often requiring the data involved in the calculation to be numerical fields. Taking the influencer recommendation scenario as an example, to determine whether an influencer matches a potential collaborator (such as the products corresponding to a brand), for products, one might use numerical parameters such as price (e.g., both around 100 yuan), category (e.g., both category 1), and user rating (e.g., both around 80 points) to calculate the similarity value according to the corresponding formula (e.g., the law of cosines, whose value ranges from [-1, 1], with the closer to 1 indicating greater similarity).

[0005] However, in real-world influencer databases, the attributes of influencers are rich and diverse. Besides the quantifiable numerical fields mentioned above, they also include numerous non-numerical fields, such as the influencer's content style (e.g., comedy, art), areas of expertise (e.g., food, beauty), and collaboration preferences (e.g., preference for long-term or short-term collaborations). These non-numerical fields are crucial for comprehensively and accurately measuring the match between influencers and potential collaborators. Traditional similarity algorithms require complex preprocessing operations when handling these non-numerical fields, such as encoding non-numerical attributes into numerical forms. This not only increases the complexity and workload of data processing but may also lead to the loss of some key semantic information during the conversion process, thus affecting the accuracy of similarity calculations and the final recommendation results.

[0006] Furthermore, relying solely on limited numerical attributes to measure matching degree makes it difficult to comprehensively characterize the characteristics of influencers and collaborators from multiple dimensions. As a result, recommendation results often only reflect the similarity of some surface features and cannot delve into deeper adaptation relationships, which greatly reduces the accuracy and effectiveness of recommendations and makes it difficult to meet the high requirements for accurate matching recommendations in actual business. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide an object matching method based on attribute alignment and feature fusion to at least partially solve the above problems.

[0008] According to a first aspect of the present invention, an object matching method based on attribute alignment and feature fusion is provided, comprising:

[0009] Obtain the attribute description text data of the first target object, and extract its features to obtain several first attribute description feature word groups;

[0010] Several first attribute description feature phrases are matched with a set attribute classification table to cluster the first attribute description feature phrases, thereby obtaining multiple first-dimensional attribute feature sequences that describe the first target object.

[0011] Vectorize each first dimension attribute feature sequence to obtain the corresponding first attribute feature vector, so as to form a feature profile of the first target object;

[0012] Obtain the attribute description text data of the second target object, and extract its features to obtain several second attribute description feature word groups;

[0013] Align several second attribute description feature phrases with several first attribute description feature phrases to cluster the second attribute description feature phrases, thereby obtaining multiple second-dimensional attribute feature sequences that describe the second target object;

[0014] Vectorize each second-dimensional attribute feature sequence to obtain the corresponding second-attribute feature vector, thereby forming a feature profile of the second target object;

[0015] Based on the feature profiles of the first target object and the second target object, the matching degree between the first target object and the second target object is calculated and compared with a set matching degree threshold to determine whether the first target object and the second target object match.

[0016] In the embodiments of the present invention, the object matching method based on attribute alignment and feature fusion has the following technical advantages:

[0017] (1) Effectively handle non-numeric fields and reduce data processing complexity.

[0018] The method provided in this invention starts by acquiring the attribute description text data of an object, and directly extracts features from the text data to obtain attribute description feature phrases. For example, for text description information such as "humorous style," "expert in the food field," and "preferring long-term cooperation" of an online celebrity, the corresponding feature phrases can be naturally extracted. Then, clustering is performed by matching with a set attribute classification table to form attribute feature sequences of different dimensions. The entire process does not require deliberately converting non-numerical fields into numerical forms, cleverly avoiding the complex preprocessing steps in traditional methods that are prone to information loss. This effectively reduces the complexity of data processing, ensures the integrity of the original semantic information of the data, and lays the foundation for subsequent accurate matching calculations.

[0019] (2) Fully integrate multi-dimensional features to improve the comprehensiveness of matching degree measurement.

[0020] The method of this invention fully considers the rich attribute information of the object. Whether it's numerical or non-numerical fields, they all undergo a series of steps such as feature extraction, clustering, and vectorization to ultimately form a complete feature profile. Taking the matching of influencers and brands as an example, the influencer's feature profile integrates vectors corresponding to multi-dimensional features such as their content creation style and areas of expertise, while the brand's profile similarly integrates multi-dimensional feature vectors such as its product characteristics and brand positioning. Calculating the matching degree based on this comprehensive feature profile that integrates multi-dimensional features allows for in-depth exploration of the deep-seated compatibility relationships between objects, comprehensively considering whether the two are a match from multiple perspectives. This greatly improves the comprehensiveness of the matching degree measurement, moving beyond superficial numerical attribute comparisons and making the recommendation results more aligned with the high requirements for accurate matching in actual business operations.

[0021] (3) Improve the accuracy of similarity calculation and optimize the recommendation effect.

[0022] The method proposed in this invention constructs feature profiles through attribute alignment and feature fusion, ensuring the accurate extraction and reasonable integration of feature information across various dimensions. When calculating the matching degree between the first and second target objects, it is based on feature profiles that fully and accurately reflect the object's characteristics. For example, when matching influencers with potential collaborative products, all kinds of key information contained in the feature profiles of both parties can be fully reflected in the matching degree calculation process, avoiding calculation errors caused by information loss or one-sided considerations. This makes the similarity calculation more accurate, enabling precise selection of truly highly matched objects for recommendation, significantly optimizing the recommendation effect, and helping to better achieve efficient utilization of influencer resources, meeting the company's influencer database's needs for accurate recommendations in marketing promotion, content dissemination, and other business scenarios.

[0023] (4) Flexible adaptation to diverse business scenarios and object attributes

[0024] The object matching method based on attribute alignment and feature fusion of this invention revolves around the processing of attribute description text data and the flexible extraction and fusion of features. It is not dependent on specific data types, such as the requirement that the data must be numerical. Any attribute information that can be conveyed through text description can be included in the matching degree calculation. This allows the method to adapt well to changes in the attributes of different types of objects in different business scenarios. Whether it's the emergence of new creative style attributes of online celebrities or the launch of new product positioning attributes by collaborating brands, these can all be integrated into the feature profile construction and matching calculation process through corresponding text descriptions, demonstrating strong flexibility and scalability. This provides a powerful technical guarantee for long-term, continuous, and accurate object matching recommendations. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0026] Figure 1 This is a schematic diagram of an object matching method based on attribute alignment and feature fusion provided in an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of an object matching device based on attribute alignment and feature fusion, provided as an embodiment of the present invention.

[0028] Figure 3 This invention provides a schematic diagram of the structure of an electronic device. Detailed Implementation

[0029] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art should fall within the protection scope of the present invention.

[0030] It should be understood that the terms "first," "second," and "third," etc., in the claims, specification, and drawings of this disclosure are used to distinguish different objects, not to describe a specific order. The terms "comprising" and "including" as used in the specification and claims of this disclosure indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or sets thereof.

[0031] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0032] Figure 1 This is a schematic diagram illustrating the process of an object matching method based on attribute alignment and feature fusion, provided in an embodiment of the present invention. Figure 1 As shown, it includes:

[0033] Obtain the attribute description text data of the first target object, and extract its features to obtain several first attribute description feature word groups;

[0034] Several first attribute description feature phrases are matched with a set attribute classification table to cluster the first attribute description feature phrases, thereby obtaining multiple first-dimensional attribute feature sequences that describe the first target object.

[0035] Vectorize each first dimension attribute feature sequence to obtain the corresponding first attribute feature vector, so as to form a feature profile of the first target object;

[0036] Obtain the attribute description text data of the second target object, and extract its features to obtain several second attribute description feature word groups;

[0037] Align several second attribute description feature phrases with several first attribute description feature phrases to cluster the second attribute description feature phrases, thereby obtaining multiple second-dimensional attribute feature sequences that describe the second target object;

[0038] Vectorize each second-dimensional attribute feature sequence to obtain the corresponding second-attribute feature vector, thereby forming a feature profile of the second target object;

[0039] Based on the feature profiles of the first target object and the second target object, the matching degree between the first target object and the second target object is calculated and compared with a set matching degree threshold to determine whether the first target object and the second target object match.

[0040] In this embodiment, the object matching method based on attribute alignment and feature fusion addresses the limitations of existing recommendation algorithms in the background art. It can effectively improve the matching accuracy, comprehensiveness, and adaptability to different business scenarios in influencer database recommendation services, and help achieve more efficient and accurate influencer resource recommendation and utilization.

[0041] Optionally, the attribute description text data of the first target object or the attribute description text data of the second target object is used as the processing object, and the following steps are performed to extract features from it to obtain the feature extraction result, wherein the feature extraction result is a plurality of first attribute description feature word groups, or a plurality of second attribute description feature word groups:

[0042] The text data corresponding to the processing object is cleaned to remove noise data, resulting in the text to be segmented.

[0043] Based on the set semantic logic, the text to be segmented is segmented to obtain several words, and a set of candidate attribute feature words is obtained based on the business domain knowledge dictionary applicable to the object to be processed. Each word belonging to the scope of the business domain knowledge dictionary is extracted as a candidate attribute feature word, so that all candidate attribute feature words form a set of candidate attribute feature words.

[0044] Calculate the term frequency and inverse document frequency of each candidate attribute feature word in the candidate attribute feature word set in the text data corresponding to the processed object;

[0045] The contribution to the description of the target object's attributes is calculated based on the term frequency and inverse document frequency of each candidate attribute feature word, wherein the target object is either the first target object or the second target object;

[0046] All candidate attribute feature words whose contribution is greater than a set contribution threshold are selected, and each selected candidate attribute feature word is taken as a valid attribute feature word, so that all valid attribute feature words form a set of valid attribute feature words;

[0047] Based on the semantic dependency relationship between the semantic logic and the business domain knowledge dictionary, the relationships of the effective attribute feature words in the effective attribute feature word set are constructed to obtain several effective attribute feature word groups.

[0048] Therefore, the above-mentioned feature extraction process has the following technical advantages:

[0049] 1. Improve data quality to ensure the accuracy of basic features.

[0050] Cleaning the text data corresponding to the target object can effectively remove noise, such as redundant punctuation marks, meaningless spaces, spelling errors, or interfering text unrelated to the target object's attribute description. Taking the text describing the attributes of online celebrities as an example, there may be some randomly entered typos or unnecessary emojis. Removing this noise yields the text to be segmented, making the data for subsequent processing cleaner and more standardized. This lays a solid foundation for accurately extracting valuable features, avoiding the extraction of incorrect or invalid features due to noise interference, and thus ensuring that the feature extraction results truly reflect the attribute situation of the target object.

[0051] The target word set is filtered using an applicable business domain knowledge dictionary, extracting only words belonging to that dictionary as candidate attribute feature words. In scenarios where influencers are matched with collaborators, such as in the beauty industry where influencers operate, the dictionary would include professional terms like "makeup tutorials," "makeup techniques," and "skincare product recommendations." This filtering method allows for precise focus on words closely related to the business, excluding words that appear in the text but have no substantial meaning in the business context. This further improves the quality of words used to construct features, ensuring that feature extraction revolves around content with business value.

[0052] 2. By comprehensively considering both word frequency and inverse document frequency, the importance of features can be accurately measured.

[0053] By statistically analyzing the word frequency and inverse document frequency (IVF) of each candidate attribute feature word in the corresponding text data of the target object, and calculating its contribution to the description of the target object's attributes, this method can comprehensively and scientifically measure the importance of each word in describing the target object's attributes. Word frequency reflects how frequently a word appears in the current object's text, reflecting its importance to some extent. However, relying solely on word frequency may lead to overemphasis on some common but overly generalized words that lack distinctiveness for specific objects. Inverse document frequency, on the other hand, considers the prevalence of the word in the entire document set (which can be understood as the total amount of relevant object text in a business context). By combining both to calculate the contribution, we can more accurately filter out words that are both frequently appearing in the current target object's text and relatively unique and more distinctive. For example, among many online celebrities, a particular celebrity's frequently mentioned and distinctive "ancient style beauty" related vocabulary can be highlighted and selected as an effective feature through this calculation. This helps to uncover feature words that truly reflect the target object's personality and key attributes, improving the accuracy of feature extraction.

[0054] 3. Construct effective attribute feature phrases to uncover deeper semantic relationships.

[0055] Based on the semantic dependency relationship between semantic logic and the business domain knowledge dictionary, relationships are constructed on the selected effective attribute feature words to obtain several effective attribute feature word groups. This operation breaks through the limitation of relying solely on single words to describe object attributes, and can uncover deep semantic connections between words, thus more comprehensively and accurately depicting the attribute characteristics of the target object. For example, in the description of internet celebrity attributes, this method can combine "ancient style" and "beauty" into the feature word group "ancient style beauty," or combine "food," "restaurant exploration," and "humorous commentary" into the feature word group "humorous commentary food exploration." The semantic information contained in these word groups is far richer than that of single words, and can more meticulously showcase the attributes of internet celebrities in terms of content creation style, areas of expertise, etc., from multiple dimensions. This allows the feature profiles subsequently constructed based on these feature word groups to more comprehensively and accurately reflect the full picture of the target object, thereby providing strong support for more accurate calculation of the matching degree with other objects and improving the effectiveness and reliability of the entire object matching process.

[0056] 4. Adapt to the characteristics of the business domain to enhance the versatility and flexibility of the methodology.

[0057] The entire feature extraction process revolves around the applicable business domain knowledge dictionary and semantic logic of the processed objects, enabling the method to adapt well to different business domains and application scenarios. Whether it's the attribute description text for different objects such as influencers, brands, and event organizers in the influencer database application scenario, or extending to text data of different objects in other fields such as e-commerce product recommendations and social user matching, as long as the corresponding business domain knowledge dictionary is constructed and the corresponding semantic logic is followed, attribute features that conform to business characteristics can be effectively extracted by following this process. This business domain-based customized processing approach enhances the method's versatility and flexibility, enabling it to be widely applied in various business scenarios that require object matching based on text attribute descriptions, meeting the diverse needs of different businesses for accurate feature extraction and object matching.

[0058] Optionally, the method further includes:

[0059] The system comprises a basic information layer, a business analysis layer, and a content feature layer. The basic information layer includes regional attribute coding, industry attribute coding, and technical data attribute coding. The business analysis layer includes market influence indicators, cooperation compatibility tags, and audience profile features. The content feature layer includes content theme tags and segmentation attribute tags.

[0060] For the basic information layer, business analysis layer, and content feature layer, index key values ​​and corresponding data pointers are assigned. The index key values ​​are used to quantify and encode the attribute features of each layer to determine the semantic and logical relationships of the basic information layer, business analysis layer, and content feature layer. The data pointers are used to record the semantic and logical relationships of the basic information layer, business analysis layer, and content feature layer.

[0061] The index key value and the corresponding data pointer are filled into the structure of the designed attribute classification table to form the attribute classification table.

[0062] Therefore, the above steps for constructing the attribute classification table have the following technical advantages:

[0063] 1. Layered architecture reflects multi-dimensional features and enhances comprehensiveness.

[0064] By constructing a basic information layer, a business analysis layer, and a content feature layer, and incorporating different types of attribute features in each layer—such as the basic information layer covering regional attribute codes, industry attribute codes, and technical data attribute codes; the business analysis layer including market influence indicators, collaboration suitability tags, and audience profile features; and the content feature layer including content theme tags and segmentation attribute tags—this approach comprehensively organizes and integrates various important information related to the target from multiple dimensions. Taking the matching scenario between influencers and collaborators as an example, this layered architecture considers not only basic information such as the influencer's region and industry, but also key elements from the business analysis level, such as their market influence and audience profile, as well as content features such as the theme and style of their content creation. This allows for a comprehensive characterization of the target, avoiding the one-sidedness caused by considering only a single dimension of the target's characteristics, and providing a rich and comprehensive feature foundation for subsequent accurate matching.

[0065] Different business scenarios and application needs may require varying degrees of focus on attributes at each level. This layered approach allows for flexible extraction and utilization of information from one or more layers, depending on the specific circumstances. For example, in business scenarios that emphasize content collaboration, more attention can be paid to the relevant attributes of the content feature layer to filter matching targets; while for collaboration scenarios that value market influence and audience fit, the characteristics of the business analysis layer can be highlighted. This layered architecture enables the entire method to better adapt to diverse business needs, enhancing its applicability and effectiveness in different application scenarios.

[0066] 2. Index keys and data pointers have clear semantic logic, enhancing relevance and operability.

[0067] By using index keys to quantify and encode the attribute features of each layer, the semantic and logical relationships between the basic information layer, the business analysis layer, and the content feature layer can be determined in a clear and standardized manner. For example, through specific encoding rules, different cities in the geographic attribute can be assigned different numerical codes as index keys, and industry attributes can also be encoded accordingly. In subsequent processing, these codes can be used to easily determine the association and hierarchical relationships between different attribute features. This allows the originally complex and diverse attribute information to be organized in an orderly manner according to a predetermined logic, facilitating efficient computer processing and analysis, and enhancing the logic and systematicity of the entire system when processing object attribute data.

[0068] Data pointers are used to record the semantic and logical relationships between different layers. They act like a navigation tool, quickly locating and tracing the relationships between attributes and features at different layers. In practical applications, when querying or using a specific attribute of an object, or analyzing its relationship with other attributes, data pointers accurately guide the system to find the corresponding information, improving the efficiency of data querying and usage. They also ensure that when performing operations such as attribute feature fusion and matching degree calculation, information can be integrated based on accurate logical relationships, avoiding errors caused by confused relationships, and further enhancing the reliability and accuracy of the entire object matching process.

[0069] 3. The attribute classification table integrates information, facilitating unified management and efficient utilization.

[0070] The attribute classification table is formed by filling the index key values ​​and corresponding data pointers into the designed attribute classification table structure, achieving structured integration of information related to each layer and attribute. All key elements related to object attribute descriptions and their logical relationships are concentrated in this table, making data resource management more orderly and convenient, avoiding the inconvenience caused by scattered data storage and management. For example, when maintaining and updating the attribute information of a large number of influencers and collaborators, the attribute classification table can clearly show the status of each object's attributes at each level and their interrelationships, facilitating unified operation and management.

[0071] As a standardized information carrier, the attribute classification table provides a unified basis and reference standard for subsequent processes such as feature extraction, clustering, vectorization, and matching degree calculation. Whether for the first or second target object, this attribute classification table can be used to accurately perform various processing operations, ensuring that different objects undergo feature analysis and matching under the same rules and framework. This guarantees the consistency and standardization of the entire object matching method, helps improve the accuracy and comparability of the final matching results, and makes the entire matching process more scientific and efficient.

[0072] Optionally, the step of matching several sets of first attribute description feature words with a set attribute classification table to cluster the first attribute description feature words to obtain multiple first-dimensional attribute feature sequences describing the first target object includes:

[0073] For each first attribute description feature word group, keyword matching is performed between it and each attribute category in the attribute classification table, and the semantic similarity between each first attribute description feature word group and the content topic tag corresponding to the matched attribute category in the matched attribute classification table is calculated using the trained semantic model.

[0074] If the similarity is greater than a set similarity threshold, then each first attribute description feature word group is assigned to the matched attribute category;

[0075] Cluster the multiple first attribute description feature word groups assigned to each attribute category to obtain multiple first dimension attribute feature sequences that describe the first target object.

[0076] Therefore, the above steps regarding matching and clustering the first attribute description feature word groups using an attribute classification table to obtain the first dimension attribute feature sequence have the following technical advantages:

[0077] 1. Achieve precise matching and improve classification accuracy by leveraging semantic similarity.

[0078] When matching the first attribute description feature phrase with the attribute categories in the attribute classification table, the matching is not simply based on surface keyword matching. Instead, a trained semantic model is used to calculate the semantic similarity between the phrase and the corresponding content topic tag. This means that the semantic relationship between words can be deeply explored, going beyond the limitations of simple text matching. For example, when matching the attribute description feature phrase "fun food sharing" of an online celebrity with the content topic tag "food experience introduction" in the food category of the attribute classification table, the semantic model can accurately determine that although the two expressions are different, they are highly related semantically, thus achieving a more accurate match. This judgment method based on semantic similarity can better capture the real relationship between feature phrases and attribute categories, avoid incorrect matching caused by differences in word expression, and significantly improve the accuracy of assigning feature phrases to appropriate attribute categories.

[0079] By setting a similarity threshold to determine whether to assign feature phrases to matched attribute categories, a quality control checkpoint is added to the matching process. Assignment is only made when the semantic similarity exceeds the threshold. This effectively filters out phrases that, while having some semantic connection, have a weak connection and may interfere with subsequent clustering and feature sequence construction. For example, in some complex business scenarios, some words may have semantic connections with multiple attribute categories, but they should only be classified into a category if they achieve a high similarity to a particular category and meet the actual business requirements. Setting a similarity threshold ensures the rationality of this assignment, guaranteeing that the feature phrases entering the subsequent clustering stage have a strong and reasonable semantic connection with the corresponding attribute category, laying the foundation for accurately constructing attribute feature sequences that reflect the characteristics of the target object.

[0080] 2. Cluster and integrate feature word groups to mine multi-dimensional attribute features.

[0081] Clustering multiple first-attribute descriptive feature phrases assigned to each attribute category can integrate phrases with similar attribute features, extracting more in-depth and comprehensive dimensional attribute features from scattered phrases. For example, in the description of internet celebrity attributes, under the food category, there may be multiple feature phrases such as "street food recommendations," "home-style food preparation sharing," and "local food exploration." Through clustering, these can be integrated into a first-dimensional attribute feature sequence about the internet celebrity's activities and style in the food field. This sequence contains far more information than a single phrase, and can more comprehensively and meticulously depict the internet celebrity's attribute characteristics in the food dimension. It helps to describe the first target object from multiple different dimensions (corresponding to different attribute categories) in an all-round way, providing strong multi-dimensional feature support for subsequently building a complete and accurate feature profile.

[0082] This clustering operation yields multiple first-dimensional attribute feature sequences, clearly revealing the multifaceted and hierarchical nature of the target object's characteristics. Different dimensional attribute feature sequences correspond to different business attribute levels, such as geographic, industry, and content style dimensions. Together, they constitute a three-dimensional description of the target object, ensuring that subsequent feature fusion and matching degree calculations with other objects fully consider the object's characteristics at each level and their interrelationships. This avoids the one-sidedness of a single-dimensional description and better reflects the reality of complex and diverse attributes in actual business scenarios, thereby improving the comprehensiveness and accuracy of the entire object matching method's grasp of the target object's characteristics.

[0083] 3. Align with business logic and semantic understanding to enhance the practicality of the method.

[0084] The entire process revolves around an attribute classification table, which is constructed based on business domain knowledge and object matching requirements. It encompasses different attribute categories and corresponding content theme tags, reflecting the logical relationships within the business scenario. By matching and clustering feature phrases according to the structure of the attribute classification table, the entire operation closely aligns with business logic and meets the actual business needs for classifying and integrating different object attributes. For example, in the e-commerce influencer and product matching scenario, the attribute category divisions and corresponding theme tag settings for products in the attribute classification table are determined based on business processes such as product sales and promotion. Processing the influencer's attribute description feature phrases according to these rules allows for better integration and matching with product attributes, enhancing the method's adaptability and practicality in real-world business applications.

[0085] By leveraging semantic models to calculate semantic similarity and subsequent clustering operations, this approach fully considers the semantic understanding of natural language, aligning more closely with how people describe and understand object attributes in real-world business scenarios. Compared to purely rule-based or simple statistical methods, this semantic-based processing better handles the diversity and flexibility of language expression, accurately extracting feature information that conforms to business semantics. This ensures that the final constructed first-dimensional attribute feature sequence truly and effectively reflects the key attributes of the first target object in the business scenario, providing a more reliable and realistic feature foundation for achieving accurate object matching. This helps improve the effectiveness and operability of the entire object matching process in real-world business environments.

[0086] Optionally, the first dimension attribute feature sequence or the second dimension attribute feature sequence is used as the processing object, and the following steps are performed to vectorize it to obtain a vectorized result, thereby forming a feature profile result. The feature profile result is the feature profile of the first target object or the feature profile of the second target object, and the vectorized result is a first attribute feature vector or a second attribute feature vector.

[0087] Based on the depth of the basic information layer, business analysis layer, and content feature layer, different embedding layers are constructed, and each embedding layer has a learnable weight matrix.

[0088] Find the index key value of each attribute description feature word group in the processing object, and vectorize it through the weight matrix of the corresponding Embedding layer to obtain the vectorized result;

[0089] Principal component analysis is performed on the vectorization result to project the vectorization result into a principal component space with a spatial dimension smaller than that of the weight matrix, thereby obtaining principal component feature vectors corresponding to the basic information layer, business analysis layer, and content feature layer.

[0090] The principal component feature vectors from the basic information layer, business analysis layer, and content feature layer are weighted and fused to obtain a weighted fused vector, which is then semantically labeled to generate the feature profile.

[0091] Therefore, the above-mentioned steps for vectorizing dimensional attribute feature sequences to form feature profiles have the following technical advantages:

[0092] 1. Construct an embedding layer based on a hierarchical structure to adapt to multi-dimensional features.

[0093] Different embedding layers are constructed based on the depth of the basic information layer, business analysis layer, and content feature layer. This hierarchical construction approach fully considers the uniqueness and differences of attribute features at different levels. For example, the relatively basic and objective information such as the regional attribute code and industry attribute code in the basic information layer, the market influence index and audience profile features in the business analysis layer which involve more business analysis, and the content theme tags and segmentation attribute tags in the content feature layer which focus on content creation characteristics, each have different semantic and data distribution characteristics. By constructing an independent embedding layer for each layer, different learning mechanisms and parameter settings can be used to specifically vectorize the feature words of each layer. This allows the vectorization results to better capture and retain the key information of each layer, improving the ability to represent complex attribute features of different dimensions.

[0094] The importance and representation of attribute information at each layer may vary across different business scenarios, and a layered embedding layer can flexibly adapt to this diversity. For example, in some influencer-brand collaborations, the focus might be more on the features of the business analysis layer to determine the match. In this case, the corresponding embedding layer can highlight the vectorization effect of these features by adjusting weights. Conversely, in collaborations that emphasize content creation, the embedding layer of the content feature layer can play a more crucial role. This flexibility allows the method to adapt to various business data with different focuses and characteristics, enhancing its overall versatility and adaptability.

[0095] 2. Utilize index key values ​​and weight matrices for vectorization to accurately represent features.

[0096] Vectorization is performed by retrieving the index keys of the feature phrases describing each attribute within the object, leveraging the index key system established during the construction of the attribute classification table. These index keys are the results of quantized encoding of the attribute features at each layer, clearly identifying the category and logical relationship of each feature phrase. This precise positioning ensures that the corresponding feature phrases are accurately mapped to the appropriate embedding layer and transformed using the correct weight matrix during vectorization. This avoids confusion and incorrect mapping during vectorization, guaranteeing that the vectorization results accurately reflect the semantics of each attribute feature phrase and its position within the overall object attribute system, providing a reliable vector foundation for subsequently building high-quality feature profiles.

[0097] Each embedding layer has a learnable weight matrix. During vectorization, the weight matrix continuously learns patterns and associations in the data, transforming feature phrases into appropriate vector representations. This allows semantically similar or related features in the vector space to exhibit corresponding correlations in terms of distance, direction, etc. For example, for different internet celebrity-related feature phrases within the same food category, after learning and vectorizing through the embedding layer's weight matrix, their positions in the vector space will be relatively close. This effectively utilizes the characteristics of the vector space to represent the intrinsic connections between attribute features, improving the expressive power of vectorization for features and facilitating more accurate analysis and utilization of these features for object matching.

[0098] 3. Principal component analysis for dimensionality reduction, highlighting key information and reducing redundancy.

[0099] Principal component analysis (PCA) of the vectorized results projects them into a principal component space with a spatial dimension smaller than that of the weight matrix, achieving dimensionality reduction. In real-world business scenarios, the original vectorized results may have high dimensionality, containing redundant information or noise. This not only increases the complexity of subsequent calculations but may also obscure truly crucial features. PCA identifies principal component dimensions with high variance and significant contribution to the overall data structure. Projecting the vectors into the principal component space formed by these key dimensions highlights the most important features and reduces interference from irrelevant or secondary information. For example, among numerous influencer attribute feature vectors, PCA can focus on dimensions that best reflect the influencer's unique style, market influence, and other key aspects. This makes subsequent operations based on these dimensionality-reduced principal component feature vectors more efficient and allows for better identification and matching analysis of the object's core features.

[0100] Principal component analysis (PCA) reduces dimensionality while preserving as many key feature relationships as possible from the original data. This means that although the dimensionality is reduced, the relative relationships between vectors and the key semantic connections implied by each dimension are largely preserved. Important information for determining the similarity or difference between objects is not lost due to dimensionality reduction. Therefore, in subsequent feature profiling and matching degree calculations, reasonable and accurate operations can still be performed based on these principal component feature vectors that retain the core feature relationships, ensuring the effectiveness and reliability of the entire object matching process.

[0101] 4. Weighted fusion and semantic annotation integrate multi-level features and enhance interpretability.

[0102] We employ weighted fusion of principal component feature vectors from different layers (basic information layer, business analysis layer, and content feature layer) to fully consider the varying importance of features at each layer to the overall object being described. In practical business scenarios, the weights of attribute features from different layers in determining object matching may differ. For example, for collaborations with certain high-end brands, the market influence feature from the business analysis layer may have a higher weight; while for creative content collaborations, the relevant attributes from the content feature layer have a greater weight. By appropriately setting weights and fusing them, we can organically integrate the key information from each layer, forming a vector representation that comprehensively reflects the multi-dimensional characteristics of the object. This avoids the one-sidedness of relying solely on features from a single layer for matching, enabling feature profiling to more comprehensively and accurately depict the object and providing richer and more effective feature basis for precise matching.

[0103] Semantic annotation of the weighted fusion vector during feature profiling endows the feature profiles with clear semantic meaning, enhancing their interpretability. In subsequent object matching and result analysis, the semantic annotations allow users to intuitively understand the business implications of each part of the feature profile and identify which attributes influence the matching score. For example, annotating one dimension to represent an influencer's "regional influence weight" and another dimension to represent "content style uniqueness" helps business personnel better understand and utilize feature profiles for decision-making. This also makes the entire object matching method more transparent, easier to understand and control in practical business applications, improving its practicality and acceptability.

[0104] Optionally, based on the depth of the basic information layer, business analysis layer, and content feature layer, different embedding layers are constructed respectively. Each embedding layer has a learnable weight matrix, including:

[0105] Assign identifiers to the embedding layers corresponding to the basic information layer, business analysis layer, and content feature layer, respectively. , , ;

[0106] For the Embedding layers corresponding to the basic information layer, business analysis layer, and content feature layer, let the number of different values ​​of the discrete features input to them be respectively... , , Set the embedded space dimension as ;

[0107] Configure the weight matrices of the Embedding layers corresponding to the basic information layer, business analysis layer, and content feature layer as follows:

[0108] For the weight matrix of the foundation information layer Embedding layer ,for Rows (corresponding to different basic information feature values) A matrix of columns, each element in the matrix Represents the first in the basic information layer The nth feature value is mapped to the embedding space. The weight parameters of the dimension;

[0109] For the weight matrix of the Embedding layer in the business analysis layer Similarly, for OK, A matrix of columns, where elements Indicates the business analysis layer The value of the feature in the embedding space is... Dimensional weights;

[0110] For the weight matrix of the content feature layer Embedding layer Its function is to map various feature labels from the content feature layer to the embedding space, with a dimension of OK, Columns, elements Indicates the business analysis layer The value of the feature in the embedding space is... The weight of the dimension.

[0111] Therefore, the above steps regarding configuring different embedding layer weight matrices have the following technical advantages:

[0112] 1. Clear labeling facilitates differentiation and management, improving operational clarity.

[0113] Assigning different identifiers to the embedding layers corresponding to the basic information layer, business analysis layer, and content feature layer clearly distinguishes each embedding layer within the entire object matching system. In subsequent feature processing, vectorization, and other related operations, these explicit identifiers allow for easy targeted calls, parameter adjustments, or result analysis of different embedding layers. For example, when it's necessary to view the vectorization effect of the basic information layer or optimize its weight matrix, the corresponding identifier can quickly locate the relevant embedding layer, avoiding confusion between different layers. This improves the clarity and management efficiency of the entire system when processing multi-layered features, facilitating more organized work based on features at each layer.

[0114] 2. Configure based on the number of feature values ​​and the dimensions of the embedding space to adapt to data characteristics.

[0115] For each embedding layer, the number of rows in the weight matrix is ​​determined based on the number of different values ​​of the discrete features input. This fully considers the diversity of values ​​of the features contained in different layers. The attribute features of the basic information layer, business analysis layer, and content feature layer often differ significantly in nature and value range. For example, the geographic attributes of the basic information layer may only have a limited number of typical values ​​(such as different city names), while the content theme tags of the content feature layer may have a richer and more diverse range of values ​​(such as various subdivided creative themes). By determining the number of rows in the weight matrix according to the number of values ​​for each layer, the size of the weight matrix can be matched with the feature richness of the corresponding layer. This allows for the reasonable allocation of mapping weights in the embedding space for each different feature value during vectorization, more accurately transforming the discrete features of each layer into appropriate vector representations, and effectively adapting to the characteristic differences of data from different layers.

[0116] A unified embedding space dimension is established, providing a standardized vector mapping space for each embedding layer. Despite the different features of each layer, vectorization within this common embedding space facilitates subsequent integration, comparison, and analysis based on a unified spatial structure. For example, in subsequent steps such as principal component analysis or weighted fusion, vectors obtained within the same-dimensional embedding space are easier to process collaboratively, avoiding complex transformations and coordination issues caused by inconsistent vector space dimensions across layers. This enhances the standardization and coherence of multi-layer feature processing throughout the object matching process, enabling features from different layers to function within a relatively unified framework, collectively contributing to accurate feature profiling and effective object matching.

[0117] 3. Clarify the meaning of the elements in the weight matrix to enhance interpretability and controllability.

[0118] This document details the meaning of each element in the weight matrix of the embedding layer, explicitly stating that each element corresponds to a weight parameter that maps a specific feature value in the corresponding layer to a specific dimension of the embedding space. This clear definition allows technical or business personnel to intuitively understand the specific meaning of each weight value when viewing and analyzing the weight matrix, and to know how a feature value affects its vector representation in the embedding space during the vectorization process. For example, in the weight matrix of the basic information layer's embedding layer, a certain element represents the weight of the regional attribute "Beijing" in a certain dimension of the embedding space. Understanding the meaning of these weights allows for a better grasp of the mapping relationship between features and vectors, enhancing the interpretability of the entire vectorization process and the weight matrix itself.

[0119] With the meaning of the weight matrix elements clear, in practical applications, if the vector representation of certain features in a certain layer does not meet expectations after vectorization, or if it is desired to strengthen or weaken the performance of certain features in the embedding space, the corresponding weight parameters can be adjusted in a targeted manner according to the specific situation. This controllability allows for optimization of the mapping relationship between features and the embedding space based on business needs and actual data performance, further improving the quality of vectorization. This ensures that the final vectors more accurately reflect the true situation of features at each layer, providing a more reliable foundation for building high-quality feature profiles and performing accurate object matching. It also enhances the flexibility and adaptability of the entire object matching method to different business scenarios and data characteristics.

[0120] Optionally, the step of finding the index key value of each attribute description feature word group in the processing object and vectorizing it through the weight matrix of the corresponding embedding layer to obtain the vectorized result includes:

[0121] Let the number of attribute description feature phrases contained in the object being processed be... For each attribute description feature phrase, it is mapped to the corresponding index key value in the basic information layer, business analysis layer, and content feature layer, respectively, and the corresponding index key value set is configured and identified as follows. , , ;

[0122] Regarding the aforementioned basic information layer, for the first... Each attribute describes a feature word group ( ), whose corresponding index key value is Through the weight matrix of the basic information layer and the embedding layer Vectorization yields a vector Take the weight matrix The index is The row vector is used as the vectorized result corresponding to the feature word group, and the vector dimension is... The calculation formula is:

[0123] For the aforementioned business analysis layer, its first Each attribute describes a feature word group ( Its index key value is Through the weight matrix of the business analysis layer and the embedding layer Vectorized vectors The calculation formula is:

[0124] For the first content feature layer Each attribute describes a feature word group ( ), index key value Through the content feature layer Embedding layer weight matrix Vectorization yields vectors The calculation formula is:

[0125] The set of vectorized results from the basic information layer is denoted as... Business analysis layer Content feature layer .

[0126] Therefore, the above steps of finding the index key value and vectorizing it through the corresponding Embedding layer weight matrix have the following technical advantages:

[0127] 1. The index key-value set configuration clearly identifies its ownership, improving processing accuracy.

[0128] To address the issue of different identifiers for the index key-value sets corresponding to each attribute description feature phrase in the object processing layer at the basic information layer, business analysis layer, and content feature layer, this approach clearly defines the affiliation of each feature phrase at different levels. In complex object attribute description scenarios, there are numerous feature phrases of different types and origins. These clear identifiers allow for quick and accurate classification into the corresponding layers, avoiding incorrect mapping or processing caused by confusion in feature phrase affiliation during vectorization. For example, when faced with numerous influencer attribute description feature phrases, the identifiers can accurately distinguish which belong to the basic information layer (e.g., region-related phrases), which belong to the business analysis layer (e.g., market influence-related phrases), and which belong to the content feature layer (e.g., content theme-related phrases). This ensures that subsequent vectorization operations are performed based on the correct layer and corresponding weight matrix, effectively improving the accuracy and orderliness of the entire vectorization process.

[0129] 2. Based on index key values ​​and weight matrices, normalized vectorization is performed to ensure the rationality of the mapping.

[0130] By mapping each attribute description feature phrase to a corresponding index key value, and then searching for the corresponding row vector in the weight matrix of each embedding layer based on the index key value as the vectorization result, a precise and standardized vectorization method is achieved. The index key value is the embodiment of the quantized encoding of the attribute features of each layer. It has a clear correspondence with the feature phrase and can be used to accurately find the row containing the weight parameters specifically configured for that feature phrase in the weight matrix. This index key-based positioning method ensures that the vectorization process strictly follows the previously constructed attribute classification and encoding system, guaranteeing that each feature phrase can be mapped into the embedding space according to predetermined rules and logic. This avoids arbitrary or incorrect vector assignment, making the conversion process from feature phrases to vectors more reasonable and scientific.

[0131] The embedding layer weight matrices of each layer are designed and learnable. Vectorized results of feature phrases are generated using the row vectors of these weight matrices, fully utilizing the learned mapping relationship between features and vectors inherent in the weight matrices. When constructing the weight matrices, factors such as the number of feature values ​​in each layer and the dimensions of the embedding space have been considered. The elements within the weight matrices represent the weight parameters of different feature values ​​in each dimension of the embedding space. Therefore, when the corresponding row vector is taken as the vectorized result based on the index key value, these carefully configured and learned weights are actually used to effectively transform feature phrases into vector representations that reasonably represent their semantics and feature information in the embedding space. This allows the vectors to accurately reflect the characteristics of the feature phrases in the corresponding layer and their correlation with other features, providing a high-quality vector foundation for subsequent feature analysis and object matching operations.

[0132] 3. Record the vectorized results hierarchically to facilitate subsequent hierarchical processing and comprehensive analysis.

[0133] The vectorization results of the basic information layer, business analysis layer, and content feature layer are recorded as different vector sets respectively. , , This hierarchical recording method facilitates subsequent targeted processing and comprehensive analysis based on the characteristics of each layer. For example, during principal component analysis, each vector set can be independently dimensionality-reduced according to the business semantics and data characteristics of its layer, revealing the most critical principal component feature vectors for each layer. During weighted fusion, it also allows for clear and reasonable fusion based on the vector sets of each layer and their corresponding importance weights, fully considering the contribution of different layers to the overall object feature description. Simultaneously, hierarchical recording facilitates the individual evaluation and comparison of the vectorization effect of each layer, helping to promptly identify and optimize potential problems in the vectorization process of a particular layer. This ensures the efficiency and accuracy of the entire process of building feature profiles and object matching based on multi-layered features, better meeting the needs of actual business operations for separately considering and comprehensively utilizing different dimensional attributes of objects.

[0134] Optionally, performing principal component analysis on the vectorized result to project the vectorized result into a principal component space with a spatial dimension smaller than that of the weight matrix, to obtain principal component feature vectors corresponding to the basic information layer, business analysis layer, and content feature layer, includes:

[0135] The vectorization results of the basic information layer are arranged into a vectorization result matrix by column. Each column is a vectorized vector, totaling... The vectorized results of the business analysis layer form a vectorized result matrix. The vectorization results of the content feature layer form a vectorization result matrix. ;

[0136] Calculate the mean of each column of the vectorized result matrix corresponding to the basic information layer, where, for the basic information layer, the mean of each column of the vectorized result matrix is... The calculation formula is: , The vectorization result of the basic information layer is represented in the first... The mean of each dimension; then the matrix After centralization, a new matrix is ​​obtained. The calculation formula is: ;

[0137] Calculate the mean of each column of the vectorized result matrix corresponding to the business analysis layer, where, for the business analysis layer, the mean of each column of the vectorized result matrix is... The calculation formula is: , The vectorization result of the business analysis layer is represented in the first... The mean of each dimension; then the matrix After centralization, a new matrix is ​​obtained. The calculation formula is: ;

[0138] Calculate the mean of each column of the vectorized result matrix corresponding to the content feature layer, where, for the content feature layer, the mean of each column of the vectorized result matrix is... The calculation formula is: , The vectorization result of the content feature layer is represented in the first... The mean of each dimension; then the matrix After centralization, a new matrix is ​​obtained. The calculation formula is: ;

[0139] Calculate the covariance matrix of the vectorized result of the basic information layer. The calculation formula is: , yes The transpose matrix and covariance matrix The variance and covariance relationships between the dimensions in the vectorization results of the basic information layer are characterized.

[0140] Calculate the covariance matrix of the vectorized results of the business analysis layer and the content feature layer. , The calculation formula is: , , yes The transpose matrix and covariance matrix The variance and covariance relationships between the dimensions in the vectorized results of the business analysis layer are characterized. yes The transpose matrix and covariance matrix The variance and covariance relationships between the dimensions in the vectorized results of the content feature layer are characterized.

[0141] For the basic information layer, perform the corresponding covariance matrix. Eigenvalue decomposition yields the corresponding eigenvalues ​​and eigenvectors, arranged in ascending order of eigenvalues. Sort the feature vectors : ;

[0142] For the business analysis layer, perform the corresponding covariance matrix. Eigenvalue decomposition yields the corresponding eigenvalues ​​and eigenvectors, arranged in ascending order of eigenvalues. Sort the feature vectors : ;

[0143] For the content feature layer, perform the corresponding covariance matrix. Eigenvalue decomposition yields the corresponding eigenvalues ​​and eigenvectors, arranged in ascending order of eigenvalues. Sort the feature vectors : ;

[0144] For the aforementioned basic information layer, the centered matrix Projecting onto the principal component space yields the principal component eigenvector matrix. The calculation formula is: ,in It is the choice of the former A matrix composed of eigenvectors: Its dimensions are ;

[0145] Before selecting the business analysis layer A matrix composed of eigenvectors : , dimension After projection, the principal component eigenvector matrix is ​​obtained. The calculation formula is: ;

[0146] Before selecting the content feature layer A matrix composed of eigenvectors : , dimension After projection, the principal component eigenvector matrix is ​​obtained. The calculation formula is: .

[0147] Optionally, the step of weightedly fusing the principal component feature vectors from the basic information layer, the business analysis layer, and the content feature layer to obtain a weighted fused vector and then semantically labeling it to generate the feature profile includes:

[0148] Let the weights assigned to the principal component feature vectors of the basic information layer, business analysis layer, and content feature layer be respectively... , , , ;

[0149] The corresponding principal component feature vectors are extracted from the principal component feature vector matrices of each layer and represented in column vector form. The basic information layer corresponds to... The business analysis layer corresponds to The content feature layer corresponds to Weighted fusion is performed to obtain the weighted fusion vector. The formulas for calculating each dimension are as follows:

[0150]

[0151] in These are the dimension indices of the feature vectors of the principal components at each layer, based on... The value range of is obtained by selecting corresponding elements from vectors at different layers and performing a weighted summation, ultimately yielding a dimension of . Weighted fusion vector It represents the first A comprehensive feature representation of an object;

[0152] The weighted fusion vectors of all objects are combined to form the final feature profile matrix. Each column corresponds to a weighted fusion vector of an object, which is represented as a feature profile vector;

[0153] Based on the dimensionality mapping relationships of each attribute feature in the attribute classification tables corresponding to the basic information layer, business analysis layer, and content feature layer after principal component analysis, the feature profiling matrix is... Semantic annotation is performed so that each dimension in the feature profile matrix has a business semantic interpretation.

[0154] Therefore, the aforementioned principal component analysis (PCA) steps offer the following technical advantages:

[0155] 1. Data organization and centralized processing

[0156] The vectorized results of each layer (basic information layer, business analysis layer, and content feature layer) are respectively constructed into matrices (e.g., , , This approach integrates scattered vectors into a unified matrix structure, making the data presentation more organized and orderly. Subsequent calculations, such as mean and covariance calculations, and other complex mathematical operations, can be performed in batches based on this standard data organization form, greatly improving the convenience and efficiency of data processing. This facilitates a holistic understanding and analysis of the data characteristics inherent in the vectorized results at each level, laying a solid foundation for uncovering the inherent patterns in the data.

[0157] By calculating the mean of each column of the vectorized result matrix at each layer and then centering it, we obtain ( , , Centering effectively eliminates the bias caused by different means across various dimensions of the data. In reality, data from different dimensions may have different average levels due to differences in units, value ranges, etc. Without centering, these mean differences can interfere with the judgment of the true relationship between dimensions and the trend of data changes. After centering, the data is distributed around the origin, which helps to highlight the relative fluctuations and interrelationships between dimensions, making subsequent analysis based on the inherent structure of the data more accurate and reliable, and creating favorable conditions for accurately extracting key feature information.

[0158] 2. Covariance matrix calculation

[0159] Calculate the covariance matrix of each layer ( , , This clearly demonstrates the variance and covariance relationships between different dimensions in the vectorized results at each layer. Variance reflects the dispersion of a single-dimensional data, indicating the inherent variability of that dimension; while covariance reveals the linear correlation between different dimensions, i.e., how a change in one dimension accompanies changes in other dimensions. For example, in the basic information layer, the covariance matrix reveals whether there is a linkage between regional and industry attribute dimensions, and the degree of their correlation. This allows for in-depth insight into the inherent structure hidden within multi-dimensional data, uncovering which dimensional combinations have a key impact on the overall data characteristics. It provides crucial clues for subsequent targeted selection of principal components and identification of the most representative feature dimensions, helping to extract the core feature relationships from complex datasets and thus more accurately grasp the key attribute characteristics of objects at each level.

[0160] The relationships between dimensions reflected in the covariance matrix can guide subsequent feature selection and dimensionality reduction operations. Based on the contribution of each dimension to data variation and the strength of their interrelationships as presented in the covariance matrix, it's possible to reasonably determine which dimensions are redundant and which play a key supporting role in the overall data features, thus deciding which dimensions to retain as principal components for dimensionality reduction. This avoids unnecessary computational complexity in subsequent processes such as object matching due to excessive irrelevant or repetitive dimensional information, while ensuring that the retained principal components can cover the main feature information of the original data to the greatest extent possible. This improves the scientific rigor and effectiveness of the entire object matching method in processing data, making object matching based on these key features more realistic and accurate.

[0161] 3. Eigenvalue decomposition and sorting

[0162] Perform eigenvalue decomposition on the covariance matrix of each layer, and sort the eigenvectors in descending order of eigenvalues ​​(e.g., ...). , , The magnitude of eigenvalues ​​(and their corresponding sorted feature vectors) is significant, representing the contribution of the principal components represented by the corresponding feature vectors to explaining the data variance. Larger eigenvalues ​​mean that the corresponding feature vectors can capture more data variations and are a major reflection of the overall data trend. In this way, we can accurately select the few most critical and representative dimensions from numerous original dimensions that describe the overall characteristics of the data. Focusing attention on these core feature dimensions for subsequent operations allows us to prioritize the key features that best reflect the differences and similarities between objects during object matching, thereby significantly improving the accuracy and efficiency of matching judgments and avoiding misjudgments caused by excessive interference from multiple important features.

[0163] The projection matrix is ​​formed by selecting the first few eigenvectors in order of their eigenvalues ​​(e.g., ...). , , The principal component space is constructed using an ordered principal component selection mechanism based on the principle of decreasing importance. This mechanism ensures that, during dimensionality reduction, the main information of the original data is preserved as much as possible while effectively removing relatively minor dimensions that have little impact on the overall features. For example, when projecting the vectorized results onto the principal component space, only the most important principal component feature vectors are selected. This not only significantly reduces the data dimensionality, lowering the burden of data storage and subsequent computation, but also ensures that the resulting principal component feature vectors retain the core feature relationships of the original data well. This provides a high-quality feature foundation for building high-quality, more discriminative feature profiles and performing accurate object matching, thus optimizing the data quality and processing efficiency of the entire object matching process.

[0164] 4. Projection to principal component space

[0165] By projecting the centered matrix onto the principal component space (e.g.) , , By obtaining the principal component eigenvector matrices for each layer, the goal of dimensionality reduction is achieved. The high-dimensional space of the original vectorized results often contains a large amount of redundant information and noise, which not only consumes a lot of storage resources but also makes subsequent data calculations and analyses complex and time-consuming, seriously affecting the efficiency of the entire object matching process, especially when dealing with large-scale object data. However, after projecting to the principal component space, the data is transformed into a lower-dimensional and more compact space, which can effectively represent the main features of the original data with fewer dimensions. This greatly reduces the amount of data and the computational complexity of subsequent processing, speeds up the calculation, and enables the entire object matching method to more efficiently handle the matching needs of a large number of objects, meeting the requirements of fast and accurate matching in real-world business scenarios.

[0166] During dimensionality reduction, because the projection is based on key principal components selected through eigenvalue decomposition, the resulting principal component feature vectors optimize feature representation while preserving the core feature relationships of the original data. These principal component feature vectors focus more on the key attribute differences and similarities of objects at different levels (basic information layer, business analysis layer, and content feature layer), enabling a more accurate characterization of the essential features of objects. Compared to the original vectorized results, their description of object features is more concise, clear, and representative. In subsequent weighted fusion of principal component feature vectors at each layer and the construction of a complete feature profile, these optimized feature vectors provide stronger support for accurately determining the matching degree between objects, thereby improving the accuracy and reliability of the entire object matching process and better serving the need for precise matching of different objects in actual business operations.

[0167] Optionally, the feature profile matrix of the first target object is: The matrix contains Each feature profile vector, used Indicates its first Each feature profile vector ( Each vector has a dimension of ), : The feature profile matrix of the second target object is: The matrix contains Each feature profile vector, used Indicates its first Each feature profile vector ( Each vector has a dimension of ), ;

[0168] The step of calculating the matching degree between the first target object and the second target object based on the feature profiles of the first target object and the second target object, and comparing it with a set matching degree threshold to determine whether the first target object and the second target object match, includes:

[0169] Based on formula , , , Calculate the first target object's first The first feature profile vector and the second target object's first feature profile vector The similarity between feature profile vectors is used to obtain Similarity matrix , The first target object is represented by the first... The first feature profile vector Each dimension element The second target object is represented by the first The first feature profile vector Each dimension element;

[0170] Based on formula Calculate the overall matching degree between the first target object and the second target object. The first target object is compared with a set matching threshold to determine whether the second target object matches the first target object.

[0171] Therefore, the above steps for calculating the matching degree between the first target object and the second target object based on feature profiling have the following technical advantages:

[0172] 1. Calculating similarity based on vector dimension elements

[0173] Comprehensive consideration of the relationships between features across all dimensions:

[0174] The formula (SimilarityMatrix(i, j)) is used to calculate the first target object's... The first feature profile vector and the second target object's first feature profile vector This method calculates the similarity between feature profile vectors, taking into full account the relationships between each dimension of the two vectors. When constructing feature profile vectors, each dimension incorporates key attribute information from different levels of the object (such as basic information, business analysis, content features, etc.), and each dimension element carries specific business semantics. Calculating similarity by covering all dimension elements means comprehensively measuring the similarity between two objects on their corresponding feature profile vectors from multiple key dimensions. This avoids the one-sidedness of focusing only on some dimensions while ignoring the impact of other important attributes on matching. It can more comprehensively and meticulously capture the fit between two objects in various aspects of features, making the similarity calculation results more valuable and more accurately reflecting the close relationship between objects based on multi-dimensional features.

[0175] The above formula essentially calculates similarity based on the dot product and magnitude of vectors, leveraging the characteristics of vector space. In vector space, the dot product reflects the relationship between vectors in terms of direction and length, while the magnitude characterizes the length attribute of the vector itself. Determining similarity by the ratio of these two factors allows for an accurate assessment of the relative positions and degree of similarity between two vectors in space. For example, if the feature profile vectors of two objects have relatively consistent directions in vector space (meaning the sum of the products of their elements is relatively large, and the product of their magnitudes is also reasonably matched), then their similarity will be high. This aligns well with the business need to measure the similarity of two objects from a multi-dimensional feature perspective, providing a reliable quantitative basis based on vector space theory for subsequent accurate judgment of object matching.

[0176] 2. Calculating the overall matching degree

[0177] The formula OverallSimilarity is used to calculate the overall matching degree between the first and second target objects, which is a comprehensive consideration of the pairwise similarity between all feature profile vectors. This is because the feature profile matrix of the first target object contains... The feature profile matrix of the second target object contains vectors. By calculating and summing the similarity between all these vector pairs and then averaging them, the overall matching degree of two objects across the multi-dimensional features covered by multiple feature profile vectors can be reflected. Different feature profile vectors may characterize the attributes of objects from different angles, stages, or aspects. By combining the similarity between them to obtain the overall matching degree, the limitations of relying solely on the similarity of a single vector pair to judge object matching are avoided. This provides a more comprehensive reflection of the overall fit between two objects, offering a more macroscopic and comprehensive basis for accurately determining whether they match.

[0178] The calculated overall matching degree is compared with a set matching degree threshold to determine whether the first target object and the second target object match, providing a clear and explicit criterion for object matching. In actual business scenarios, different business needs and application scenarios may have different requirements for the degree of matching. By pre-setting an appropriate matching degree threshold, it is possible to flexibly determine which object combinations meet business requirements and can be considered a match based on specific circumstances. This threshold-based judgment mechanism makes the entire matching process more in line with actual business logic, accurately filtering out object pairs that meet specific matching criteria. This helps improve the effectiveness and applicability of object matching results in actual business applications, meeting diverse business needs. For example, in the scenario of matching influencers with collaborators, different thresholds can be set according to the degree of cooperation required to select suitable matching combinations.

[0179] Based on the above embodiments, this application also provides an object matching method, which includes:

[0180] Retrieve the attribute description data of the first and second target objects to be matched;

[0181] The attribute description data of each target object are processed by feature extraction to obtain multiple attribute feature groups for each target object.

[0182] The attribute feature groups of each target object are classified and clustered according to preset rules to form a corresponding multi-dimensional attribute feature sequence for each target object;

[0183] Vectorization operations are performed on the multi-dimensional attribute feature sequences of each target object to construct a feature profile of each target object;

[0184] Based on the feature profiles of the first target object and the second target object, the matching degree between the two is calculated and compared with the preset matching degree judgment standard to determine whether the first target object and the second target object match each other.

[0185] Figure 2 This is a schematic diagram of an object matching device based on attribute alignment and feature fusion, provided as an embodiment of the present invention. Figure 2 As shown, it includes:

[0186] The first program unit is used to obtain the attribute description text data of the first target object and extract its features to obtain several first attribute description feature word groups;

[0187] The second program unit is used to match several first attribute description feature word groups with a set attribute classification table to cluster the first attribute description feature word groups and obtain multiple first-dimensional attribute feature sequences that describe the first target object.

[0188] The third program unit is used to vectorize each first dimension attribute feature sequence to obtain the corresponding first attribute feature vector, so as to form a feature profile of the first target object.

[0189] The fourth program unit is used to acquire the attribute description text data of the second target object and extract its features to obtain several second attribute description feature word groups;

[0190] The fifth program unit is used to align several second attribute description feature word groups with several first attribute description feature word groups to cluster the second attribute description feature word groups to obtain multiple second-dimensional attribute feature sequences that describe the second target object.

[0191] The sixth program unit is used to vectorize each second dimension attribute feature sequence to obtain the corresponding second attribute feature vector, so as to form a feature profile of the second target object;

[0192] The seventh program unit is used to calculate the matching degree between the first target object and the second target object based on the feature profile of the first target object and the feature profile of the second target object, and compare it with a set matching degree threshold to determine whether the first target object and the second target object match.

[0193] The above Figure 2 In the embodiments, an exemplary explanation of the technical processing procedures for the execution of each program unit can be found above. Figure 1 The records.

[0194] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. For example... Figure 3 As shown, the electronic device includes a memory and a processor. The memory stores a computer-executable program, and the processor runs the computer-executable program to perform the following steps:

[0195] Obtain the attribute description text data of the first target object, and extract its features to obtain several first attribute description feature word groups;

[0196] Several first attribute description feature phrases are matched with a set attribute classification table to cluster the first attribute description feature phrases, thereby obtaining multiple first-dimensional attribute feature sequences that describe the first target object.

[0197] Vectorize each first dimension attribute feature sequence to obtain the corresponding first attribute feature vector, so as to form a feature profile of the first target object;

[0198] Obtain the attribute description text data of the second target object, and extract its features to obtain several second attribute description feature word groups;

[0199] Align several second attribute description feature phrases with several first attribute description feature phrases to cluster the second attribute description feature phrases, thereby obtaining multiple second-dimensional attribute feature sequences that describe the second target object;

[0200] Vectorize each second-dimensional attribute feature sequence to obtain the corresponding second-attribute feature vector, thereby forming a feature profile of the second target object;

[0201] Based on the feature profiles of the first target object and the second target object, the matching degree between the first target object and the second target object is calculated and compared with a set matching degree threshold to determine whether the first target object and the second target object match.

[0202] The above Figure 3 In the embodiments, an exemplary explanation of the technical processing procedures for each step can be found above. Figure 1 The records.

[0203] The above embodiments are only used to illustrate the embodiments of the present invention and are not intended to limit the embodiments of the present invention. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of the present invention. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of the present invention, and the patent protection scope of the embodiments of the present invention should be defined by the claims. The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or physical entities, or by products with certain functions.

[0204] For ease of description, the above apparatus is described by dividing it into various functional units. Of course, in implementing this invention, the functions of each unit can be implemented in one or more software and / or hardware components.

[0205] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

Claims

1. An object matching method based on attribute alignment and feature fusion, characterized in that, The method comprises the following steps: obtain attribute description text data of a first target object and perform feature extraction thereon to obtain a plurality of first attribute description feature phrases; match the plurality of first attribute description feature phrases with a set attribute classification table to cluster the first attribute description feature phrases and obtain a plurality of first-dimensional attribute feature sequences describing the first target object; vectorize each of the first-dimensional attribute feature sequences to obtain a corresponding first attribute feature vector to form a feature portrait of the first target object; obtain attribute description text data of a second target object and perform feature extraction thereon to obtain a plurality of second attribute description feature phrases; align the plurality of second attribute description feature phrases with the plurality of first attribute description feature phrases to cluster the second attribute description feature phrases and obtain a plurality of second-dimensional attribute feature sequences describing the second target object; vectorize each of the second-dimensional attribute feature sequences to obtain a corresponding second attribute feature vector to form a feature portrait of the second target object; based on the feature portrait of the first target object and the feature portrait of the second target object, calculate the matching degree between the first target object and the second target object and compare it with a set matching degree threshold to determine whether the first target object and the second target object match; wherein the method further comprises: constructing a basic information layer, a business analysis layer, and a content feature layer, the basic information layer comprising regional attribute codes, industry attribute codes, and technical data attribute codes, the business analysis layer comprising market influence indicators, cooperation adaptation degree labels, and audience portrait features, and the content feature layer comprising content theme labels and segmentation attribute labels; allocating index key values and corresponding data pointers for the basic information layer, the business analysis layer, and the content feature layer, the index key values being used to quantitatively encode the attribute features of each layer to determine the semantic logical relationships of the basic information layer, the business analysis layer, and the content feature layer, and the data pointers being used to record the semantic logical relationships of the basic information layer, the business analysis layer, and the content feature layer; filling the index key values and the corresponding data pointers into the structure of the designed attribute classification table to form the attribute classification table. 2.The object matching method based on attribute alignment and feature fusion according to claim 1, characterized in that, taking the attribute description text data of the first target object or the attribute description text data of the second target object as a processing object and performing the following steps to perform feature extraction thereon to obtain a feature extraction result, the feature extraction result being a plurality of first attribute description feature phrases or a plurality of second attribute description feature phrases: clean the text data corresponding to the processing object to remove noise data therefrom to obtain a text to be segmented; Based on the set semantic logic, the text to be segmented is processed to obtain a plurality of words, and based on the business domain knowledge dictionary applicable to the processing object, candidate attribute feature words are screened from the set of words to be screened, and each word within the range of the business domain knowledge dictionary is extracted as a candidate attribute feature word, so that all candidate attribute feature words form a candidate attribute feature word set; The word frequency and inverse document frequency of each candidate attribute feature word in the processing object corresponding text data are counted; According to the word frequency and inverse document frequency of each candidate attribute feature word, the contribution degree of the target object attribute description is calculated, and the target object is the first target object or the second target object; All candidate attribute feature words with a contribution degree greater than a set contribution degree threshold are screened out, and each screened candidate attribute feature word is used as an effective attribute feature word, so that all effective attribute feature words form an effective attribute feature word set; Based on the semantic dependency driven relationship between the semantic logic and the business domain knowledge dictionary, the relationship between the effective attribute feature words in the effective attribute feature word set is constructed to obtain a plurality of effective attribute feature word groups. 3.The object matching method based on attribute alignment and feature fusion according to claim 1, characterized in that, The first attribute description feature word group is matched with the set attribute classification table to cluster the first attribute description feature word group, and a plurality of first dimension attribute feature sequences for describing the first target object are obtained, including: For each first attribute description feature word group, keyword matching is performed with each attribute category in the attribute classification table, and a semantic similarity between the each first attribute description feature word group and the content theme label corresponding to the matched attribute category in the matched attribute classification table is calculated using a trained semantic model; If the similarity is greater than a set similarity threshold, the each first attribute description feature word group is assigned to the matched attribute category; A plurality of first attribute description feature word groups assigned to each attribute category are clustered to obtain a plurality of first dimension attribute feature sequences for describing the first target object.

4. The object matching method based on attribute alignment and feature fusion according to claim 3, characterized in that, The first dimension attribute feature sequence or the second dimension attribute feature sequence is taken as a processing object, and the following steps are performed to vectorize it to obtain a vectorization result to form a feature image result, the feature image result being a feature image of the first target object or a feature image of the second target object, the vectorization result being a first attribute feature vector or a second attribute feature vector: Different Embedding layers are constructed according to the depth of the basic information layer, the business analysis layer, and the content feature layer, and each Embedding layer has a learnable weight matrix; The index key value of each attribute description feature word group in the processing object is found, and the weight matrix of the corresponding Embedding layer is vectorized to obtain a vectorization result; perform principal component analysis on the vectorization result to project the vectorization result into a principal component space with a spatial dimension less than a dimension of the weight matrix, to obtain principal component feature vectors corresponding to the basic information layer, the business analysis layer, and the content feature layer; perform weighted fusion on the principal component feature vectors corresponding to the basic information layer, the business analysis layer, and the content feature layer, to obtain a weighted fusion vector and perform semantic labeling thereon, to generate the feature portrait.

5. The object matching method based on attribute alignment and feature fusion according to claim 4, characterized in that, different Embedding layers are constructed according to depths of the basic information layer, the business analysis layer, and the content feature layer, and each Embedding layer has a learnable weight matrix, including: Assign an identity to the Embedding layer corresponding to the basic information layer, business analysis layer, and content feature layer, respectively , , ; For the Embedding layer corresponding to the basic information layer, the business analysis layer, and the content feature layer, suppose that the number of different values of the discrete features inputted by the Embedding layer is , , , and the dimension of the embedding space is set to . The weight matrix of the Embedding layer corresponding to the basic information layer, the business analysis layer, and the content feature layer is configured in the following manner: For the weight matrix of the foundational information layer Embedding layer ,for OK, A matrix of columns, each element in the matrix Represents the first in the basic information layer The nth feature value is mapped to the embedding space. The weight parameters of the dimension; For the weight matrix of the Embedding layer in the business analysis layer Similarly, for OK, A matrix of columns, where elements Indicates the business analysis layer The value of the feature in the embedding space is... Dimensional weights; The content feature layer Embedding layer weight matrix , which maps each feature label of the content feature layer to an embedding space with dimension rows, columns, and element represents the weight of the th feature value of the business analysis layer in the th dimension of the embedding space.

6. The object matching method based on attribute alignment and feature fusion according to claim 4, characterized in that, The index key value of each attribute description feature word group in the processing object is found, and vectorization is performed on the weight matrix of the corresponding Embedding layer to obtain a vectorization result, including: The number of attribute description feature word groups contained in the processing object is set as For each attribute description feature word group, it is mapped to a corresponding index key value. The index key value set corresponding to the basic information layer, the business analysis layer, and the content feature layer is configured as 、 、 ; Regarding the aforementioned basic information layer, for the first... Each attribute describes a feature word group ( ), whose corresponding index key value is Through the weight matrix of the basic information layer and the embedding layer Vectorization yields a vector Take the weight matrix The index is The row vector is used as the vectorized result corresponding to the feature word group, and the vector dimension is... The calculation formula is: ; For the aforementioned business analysis layer, its first Each attribute describes a feature word group ( Its index key value is Through the weight matrix of the business analysis layer and the embedding layer Vectorized vectors The calculation formula is: ; For the first content feature layer Each attribute describes a feature word group ( ), index key value Through the content feature layer Embedding layer weight matrix Vectorization yields vectors The calculation formula is: ; The vectorization result vector set of the basic information layer is denoted as , the vectorization result vector set of the business analysis layer is denoted as , and the vectorization result vector set of the content feature layer is denoted as .

7. The object matching method based on attribute alignment and feature fusion according to claim 4, characterized in that, The principal component analysis is performed on the vectorization result to project the vectorization result into a principal component space with a spatial dimension less than a dimension of the weight matrix, to obtain principal component feature vectors corresponding to the basic information layer, the business analysis layer, and the content feature layer, including: The vectorization result of the basic information layer is grouped into a vectorization result matrix by column wherein each column is a vectorized vector, and there are columns, and the vectorization result of the business analysis layer is grouped into a vectorization result matrix , and the vectorization result of the content feature layer is grouped into a vectorization result matrix ; Calculate the mean of each column of the vectorized result matrix corresponding to the basic information layer, where, for the basic information layer, the mean of each column of the vectorized result matrix is... The calculation formula is: , The vectorization result of the basic information layer is represented in the first... The mean of each dimension; then the matrix A new matrix is ​​obtained by centralizing the matrix. The calculation formula is: ; calculating a mean value of each column of the vectorization result matrix corresponding to the business analysis layer, wherein the mean value of each column of the vectorization result matrix corresponding to the business analysis layer The calculation formula is: , represents the mean value of the business analysis layer vectorization result in the first dimension; then the matrix is centered to obtain a new matrix , and the calculation formula is: ; Calculate the mean of each column of the vectorized result matrix corresponding to the content feature layer, where, for the content feature layer, the mean of each column of the vectorized result matrix is... The calculation formula is: , The vectorization result of the content feature layer is represented in the first... The mean of each dimension; then the matrix A new matrix is ​​obtained by centralizing the matrix. The calculation formula is: ; a covariance matrix of the vectorization result of the base information layer , the calculation formula is: , is the transpose matrix of covariance matrix , representing the variance and covariance relationship between each dimension in the vectorization result of the base information layer; a covariance matrix of the vectorization results of the business analysis layer, the content feature layer , , the calculation formula is: , , is a transpose matrix of , the covariance matrix characterizes the variance and covariance relationship between each dimension in the vectorization result of the business analysis layer; is a transpose matrix of , the covariance matrix characterizes the variance and covariance relationship between each dimension in the vectorization result of the content feature layer; performing eigenvalue decomposition to obtain corresponding eigenvalues and eigenvectors, and arranging the eigenvalues in descending order performing eigenvalue decomposition to obtain corresponding eigenvalues and eigenvectors, and arranging the eigenvalues in descending order arranging the eigenvectors in descending order : ; performing eigenvalue decomposition to obtain corresponding eigenvalues and eigenvectors, and ordering the eigenvalues from large to small performing eigenvalue decomposition to obtain corresponding eigenvalues and eigenvectors, and ordering the eigenvalues from large to small ordering the eigenvectors ;​ performing eigenvalue decomposition to obtain corresponding eigenvalues and eigenvectors, and arranging the eigenvalues in descending order performing eigenvalue decomposition to obtain corresponding eigenvalues and eigenvectors, and arranging the eigenvalues in descending order arranging the eigenvectors in descending order : ; For the base information layer, the centralized matrix is projected to the principal component space to obtain a principal component feature vector matrix , and the calculation formula is: wherein is a matrix composed of the first selected feature vectors: , and the dimension is ; For the business analysis layer selection before one feature vector matrix : , dimension , after projection, the principal component feature vector matrix , the calculation formula is: ; For the content feature layer selection before The matrix composed of the feature vectors : The dimension is The principal component feature vector matrix is obtained after projection The calculation formula is: .

8. The object matching method based on attribute alignment and feature fusion according to claim 4, characterized in that, The principal component feature vectors corresponding to the basic information layer, the business analysis layer, and the content feature layer are weighted and fused to obtain a weighted fusion vector and perform semantic labeling thereon, to generate the feature portrait, including: The weights assigned to the principal component feature vectors of the basic information layer, the business analysis layer, and the content feature layer are respectively , , , ; The corresponding principal component feature vectors are taken out from the principal component feature vector matrix of each layer to be expressed in column vector form, the corresponding of the basic information layer is , the corresponding of the business analysis layer is , and the corresponding of the content feature layer is ; The weighted fusion vector is obtained by performing weighted fusion The calculation formula of each dimension is: ； wherein are the dimension index of the principal component feature vector of each layer respectively, according to the value range of the corresponding elements in the vector of different layers are selected to weighted sum, finally get the weighted fusion vector with dimension , which represents the comprehensive feature representation of the th object; The weighted fusion vectors of all objects are combined to form a final feature portrait matrix Each column corresponds to a weighted fusion vector of an object as a feature portrait vector Based on the dimensionality mapping relationships of each attribute feature in the attribute classification tables corresponding to the basic information layer, business analysis layer, and content feature layer after principal component analysis, the feature profiling matrix is... Semantic annotation is performed so that each dimension in the feature profile matrix has a business semantic interpretation.

9. The object matching method based on attribute alignment and feature fusion according to claim 4, characterized in that, The feature portrait matrix of the first target object is , which contains feature portrait vectors, denoted as , where is the th feature portrait vector of the first target object, and the dimension of each vector is : ; the feature portrait matrix of the second target object is , which contains feature portrait vectors, denoted as , where is the th feature portrait vector of the second target object, and the dimension of each vector is ; The matching degree between the first target object and the second target object is calculated based on the feature portrait of the first target object and the feature portrait of the second target object, and is compared with a set matching degree threshold, to determine whether the first target object and the second target object match, including: Based on formula , , , Calculate the first target object's first The first feature profile vector and the second target object's first feature profile vector The similarity between feature profile vectors is used to obtain Similarity matrix , The first target object is represented by the first... The first feature profile vector Each dimension element The second target object is represented by the first The first feature profile vector Each dimension element; Based on the formula , the overall matching degree between the first target object and the second target object is calculated to compare with the set matching degree threshold to determine whether the first target object matches the second target object.

Images