A feature correlation-based duplicate rental item identification method, device and medium
Patent Information
- Application Number
- CN202611027933.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-10
- Publication Date
- 2026-10-13
AI Technical Summary
[0003]目前，租赁物的查询与重复识别主要存在以下问题：一是租赁物属性特征提取不准确且不完整，租赁物信息通常分散于租赁合同、资产清单、验收单等不同文档中，且有些以表格形式存在，难以完整解析导致属性值提取有误或遗漏，同时，不同文本中同一租赁物的属性名称表述不一致、属性维度不同等，使得传统的关键词检索方式不易直接有效地进行比对，难以精准判定分散在不同文档中的租赁物是否为同一实体；二是缺乏高效的重复识别机制，现有方法多依赖单一标识字段进行匹配或多属性比对，但缺乏对多特征匹配结果的有效质量判定策略，导致识别准确率低，需人工复核，重复识别效率较低
本发明提供了一种基于特征关联的重复租赁物识别方法，首先从历史租赁文本中提取历史租赁物标识和每一历史租赁物的历史属性特征集，并构建历史属性特征关联索引表，为后续租赁物的重复租赁识别奠定了数据基础；从租赁物查询文本中提取目标租赁物标识和目标属性特征集，然后为每一目标属性特征匹配相关联的历史属性特征，并基于匹配的历史属性特征的名称和对应的历史属性值，计算历史属性特征与目标属性特征之间的关联强度，通过属性特征语义相似度与属性值向量相似度的双重匹配，能够实现跨文本的特征精准关联与鲁棒匹配，为后续综合评估租赁物重复性提供了准确可靠的量化依据；再将关联强度按匹配出的历史属性特征所属的历史租赁物进行聚合，计算每个历史租赁物与目标租赁物的综合关联得分，识别目标租赁物与对应最高综合关联得分的历史租赁物是否为同一租赁物，在计算综合关联得分时，实现了多维度参数融合的量化评估，使得到的综合关联得分更可靠，从而能够实现对多源租赁文本中同一租赁物的重复识别，并显著提升了租赁物的重复识别效率与准确性。
Smart Images
Figure CN122903980A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rental inquiry technology, and in particular to a method, device and medium for identifying duplicate rental items based on feature association. Background Technology
[0002] In business scenarios such as financial leasing and asset management, when enterprises apply for new leases or add assets to their inventory, they need to compare the leased assets currently pending with the historically registered leased assets to determine whether the leased asset has been leased or is still in inventory, thereby preventing risks such as multiple leases of the same asset or double renting.
[0003] Currently, the main problems with querying and identifying duplicates of leased assets are as follows: First, the extraction of attribute features of leased assets is inaccurate and incomplete. Leased asset information is usually scattered across different documents such as lease contracts, asset lists, and acceptance forms, and some are in tabular form, making it difficult to parse completely, leading to errors or omissions in attribute value extraction. At the same time, the attribute names of the same leased asset in different texts are inconsistent, and the attribute dimensions are different, making it difficult for traditional keyword retrieval methods to directly and effectively compare and accurately determine whether the leased assets scattered in different documents are the same entity. Second, there is a lack of efficient duplicate identification mechanisms. Existing methods mostly rely on matching single identifier fields or comparing multiple attributes, but lack effective quality judgment strategies for multi-feature matching results, resulting in low identification accuracy, requiring manual review, and low efficiency in duplicate identification. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention provides a method, device, and medium for identifying duplicate leased items based on feature association, which can accurately identify the same leased item in different lease texts and improve the efficiency of duplicate leased item identification.
[0005] According to a first aspect of the present invention, a method for identifying duplicate leased items based on feature association is provided, comprising the following steps: S1. Extract the historical lease item identifier and the historical attribute feature set of each historical lease item from the obtained historical lease text, and construct a historical attribute feature association index table based on the historical attribute feature set.
[0006] S2, when receiving the rental item query text input by the user, extract the target rental item identifier and the target attribute feature set of each target rental item from the rental item query text.
[0007] S3. For each target leased item, use any target attribute feature in the target attribute feature set of the target leased item as the query key, match at least one historical attribute feature from the historical attribute feature association index table, and calculate the association strength between each matched historical attribute feature and the target attribute feature based on the name of the historical attribute feature and the corresponding historical attribute value.
[0008] S4. Aggregate the association strengths corresponding to all target attribute features according to the historical leased items to which the matched historical attribute features belong, and calculate the comprehensive association score between each historical leased item and the target leased item.
[0009] S5. When the highest comprehensive correlation score is greater than the preset judgment threshold, the target leased item is determined to be the same leased item as the historical leased item with the highest comprehensive correlation score, and the recognition result is output.
[0010] According to a second aspect of the present invention, a non-transitory computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or the at least one program being loaded and executed by a processor to implement the above-described feature association-based method for identifying duplicate leased items.
[0011] According to a third aspect of the present invention, an electronic device is provided, including a processor and the aforementioned non-transitory computer-readable storage medium.
[0012] The present invention has at least the following beneficial effects: This invention provides a method for identifying duplicate rentals based on feature association. First, it extracts historical rental item identifiers and historical attribute feature sets for each historical rental item from historical rental texts, and constructs a historical attribute feature association index table, laying the data foundation for subsequent duplicate rental identification. Next, it extracts target rental item identifiers and target attribute feature sets from rental item query texts. Then, it matches each target attribute feature with associated historical attribute features, and calculates the association strength between the historical attribute features and target attribute features based on the names and corresponding historical attribute values of the matched historical attribute features. This is achieved through a dual matching method using semantic similarity of attribute features and similarity of attribute value vectors. The system enables precise feature association and robust matching across texts, providing accurate and reliable quantitative basis for subsequent comprehensive evaluation of the repetition of leased items. Furthermore, the association strength is aggregated according to the historical leased items to which the matched historical attribute features belong, and a comprehensive association score is calculated between each historical leased item and the target leased item. This identifies whether the target leased item and the historical leased item with the highest comprehensive association score are the same leased item. In calculating the comprehensive association score, a multi-dimensional parameter fusion quantitative evaluation is implemented, making the obtained comprehensive association score more reliable. This enables the repeated identification of the same leased item in multi-source lease texts and significantly improves the efficiency and accuracy of repeated leased item identification. Attached Figure Description
[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0014] Figure 1 A flowchart of a feature-association-based method for identifying duplicate leased items provided in an embodiment of the present invention. Detailed Implementation
[0015] The technical solutions of the embodiments of the present invention will be clearly and completely 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 without creative effort are within the scope of protection of the present invention.
[0016] This invention provides a method for identifying duplicate leased items based on feature association, such as... Figure 1 As shown, the method includes the following steps: S1. Extract the historical lease item identifier and the historical attribute feature set for each historical lease item from the acquired historical lease text, and construct a historical attribute feature association index table based on the historical attribute feature set. This can be understood as follows: historical lease text is text recording historical lease information, including but not limited to the historical lease item identifier, lease time, lessor information, and historical attribute features of the historical lease item. Historical attribute features include, but are not limited to, the leased equipment name, model, manufacturer, serial number, size, weight, color, purchase price, and lease term.
[0017] Specifically, when extracting historical leased item identifiers, for identifiers with a fixed format, such as equipment number: SB-XXXX, a predefined regular expression is used to extract identifiers that conform to the format in batches. For identifiers without a fixed format, such as contract number + serial number, a pre-trained BERT-NER model is used to label the leased item identifier entities in the historical lease text, achieving accurate extraction. If there are missing identifiers, such as some text not labeling the equipment number, a temporary unique identifier is generated using the document number + equipment name.
[0018] In addition, for tabular text, a table parsing algorithm is used to extract the content of table cells, merge information from cells spanning multiple rows or columns, and extract the data in key-value pairs of attribute name and attribute value to ensure that each attribute feature can be traced back to the original lease text. The PDFBox algorithm can be used for table parsing.
[0019] As described above, by constructing a historical attribute feature association index table, the attribute features of historical leased items are efficiently summarized and quickly retrieved, laying a data foundation for the identification of repeated leases of leased items in the future, greatly improving the query efficiency during online identification, and reducing the computational overhead of comparing large amounts of data.
[0020] Specifically, the construction of the historical attribute feature association index table based on the historical attribute feature set includes the following steps: S101, for each historical leased item, standardize each historical attribute feature and its corresponding historical attribute value into a key-value pair format of standard attribute feature name - standardized attribute value; wherein, the standardized attribute value is obtained by processing the historical attribute value according to the attribute type of the historical attribute feature. For example, equipment model and product model are standardized to model, whose attribute type is text, and operations such as special symbol removal, prefix redundancy removal, and full-width / half-width conversion are performed on the historical attribute value of this type of attribute feature; weight and gross weight are standardized to weight, whose attribute type is numeric, and the units of the historical attribute value of this type of attribute feature are standardized, and the number of significant digits is determined; in addition, for date type attribute types, a unified format is used; for enumerated type attribute types, such as color, a unified classification code is used.
[0021] S102, using the key-value pairs as index keys, an inverted index table pointing to historical leased item identifiers is established as a historical attribute feature association index table. In specific implementations, standardized attribute values of the same attribute type for different historical leased items can be pre-calculated for attribute value similarity after vector transformation, and a similarity matrix can be constructed. The similarity matrix is used to accelerate the calculation of attribute value similarity when retrieving associated historical attribute features.
[0022] As described above, by standardizing historical attribute features into a unified key-value pair format, the problem of inconsistent attribute feature names and attribute value formats in different texts is solved. At the same time, by building an inverted index table with key-value pairs as index keys, it is possible to quickly retrieve historical leased property features and improve the matching efficiency of subsequent attribute features.
[0023] S2, when the user inputs the rental item query text, extract the target rental item identifier and the target attribute feature set of each target rental item from the rental item query text; it can be understood that the extraction method of the target rental item identifier and the target attribute feature set is the same as the extraction method of the historical rental item identifier and the historical attribute feature set, and will not be repeated here.
[0024] Specifically, the rental property query text refers to the text in which a user queries the historical rental information of a target rental property. By querying the historical rental information of a target rental property, it is possible to determine whether the target rental property is currently under lease, thus preventing duplicate leasing.
[0025] S3. For each target leased item, use any target attribute feature in the target attribute feature set of the target leased item as the query key, match at least one historical attribute feature from the historical attribute feature association index table, and calculate the association strength between each matched historical attribute feature and the target attribute feature based on the name of the historical attribute feature and the corresponding historical attribute value.
[0026] Specifically, step S3 includes the following steps: S301, for any target attribute feature, calculate the semantic similarity between the target attribute feature and each historical attribute feature in the historical attribute feature association index table. Historical attribute features with semantic similarity greater than a preset similarity threshold are used as the matched historical attribute features. In specific implementations, the Word2Vec model can be used to perform vector transformation between the target attribute feature and the historical attribute feature, and cosine similarity can be used to calculate semantic similarity. Those skilled in the art can set the preset similarity threshold according to actual needs, which will not be elaborated here.
[0027] S302, for each matched historical attribute feature, look up the historical attribute value corresponding to the historical attribute feature in the historical attribute feature association index table, and calculate the vector similarity between the found historical attribute value and the target attribute value of the target attribute feature according to the corresponding calculation method based on the attribute type.
[0028] Specifically, for numerical attribute types, such as weight, size, and purchase price, the vector similarity S1 is calculated using the following formula: S1=1-|V p -V h | / (max(V p V h )+ε), where V p V is the target attribute value. h The value is the historical attribute value, and ε is a preset minimum positive number to prevent the denominator from being 0.
[0029] For short text attribute types, edit distance normalization is used to calculate vector similarity; for long text attribute types, such as specification notes, the Jaccard similarity formula is used to calculate vector similarity. The word set of the long text is obtained using the jieba word segmentation tool. Those skilled in the art are familiar with the specific calculation methods of edit distance normalization and Jaccard similarity, and will not elaborate further here.
[0030] For enumerated attribute types, such as color and device status, if the historical attribute value is the same as the target attribute value, the vector similarity is 1. If there is a preset approximate mapping, such as dark red and red being an approximate mapping, the vector similarity is 0.5; otherwise, it is 0.
[0031] For date-type attribute types, the vector similarity S2 is calculated as follows: S2=1-min(1,D / (D max +1)), where D is the absolute value of the difference in the number of days between the historical attribute value and the target attribute value, D max This is the preset maximum allowable number of days difference.
[0032] S303, calculate the association strength between each matched historical attribute feature and the target attribute feature by performing a weighted sum based on the semantic similarity and vector similarity.
[0033] Preferably, the preset weight for vector similarity is greater than the preset weight for semantic similarity. This setting aligns with the situation where attribute value matching is more discriminative than attribute feature matching.
[0034] The above approach first filters out historical attribute features that are significantly similar to the attribute features, and then uses targeted algorithms to calculate the similarity of attribute values based on the attribute type. This effectively solves the heterogeneity problems such as inconsistent attribute feature names for the same leased item in different texts and inconsistent format of multiple types of attribute values. By dual matching of semantic similarity of attribute features and similarity of attribute value vectors, and weighted fusion to obtain the association strength, it can achieve accurate feature association and robust matching across texts, providing an accurate and reliable quantitative basis for subsequent comprehensive evaluation of the repeatability of leased items.
[0035] S4. Aggregate the association strengths corresponding to all target attribute features according to the historical leased items to which the matched historical attribute features belong, and calculate the comprehensive association score between each historical leased item and the target leased item.
[0036] In one specific embodiment, step S4 includes the following steps: S401, for any historical leased property, when multiple target attribute features match the same historical attribute feature of the historical leased property, retain the matching pair between the target attribute feature with the highest correlation strength and the historical attribute feature as the valid attribute feature pair associated with the historical leased property; this can be understood as: one historical attribute feature matches only one target attribute feature.
[0037] S402, obtain all valid attribute feature pairs associated with the historical leased property, and form a set of valid attribute feature pairs corresponding to the historical leased property.
[0038] S403, Calculate the comprehensive association score between the historical leased assets and the target leased assets based on the set of valid attribute features corresponding to the historical leased assets; wherein, the comprehensive association score meets the following conditions: Where S is the comprehensive correlation score, n is the number of valid attribute feature pairs in the set of valid attribute feature pairs corresponding to historical leased items, and r is the number of valid attribute feature pairs in the set of valid attribute feature pairs. i w represents the correlation strength between the target attribute feature and the historical attribute feature in the i-th valid attribute feature pair. i is the preset importance weight of the historical attribute features in the i-th effective attribute feature pair, c is the target attribute feature coverage corresponding to the effective attribute feature pair set, and p is the preset feature conflict penalty coefficient.
[0039] Furthermore, the target attribute feature coverage rate is the number of different target attribute features in the effective attribute feature pair set divided by the total number of non-missing attribute features corresponding to the target lease item; wherein, the non-missing attribute features refer to the attribute features that actually have values in the target attribute feature set of the target lease item. This setting solves the problem of low scores caused by a small number of target attribute features and sparse data.
[0040] Specifically, when the same historical attribute feature matches multiple different attribute values, it is determined that there is a feature conflict. The preset feature conflict penalty coefficient is 1 when there is no conflict, and a positive number less than 1, such as 0.5, when there is a conflict. In the actual implementation, if a unique identifier attribute feature matches multiple different attribute values, the historical leased item is directly excluded, and it is considered that it is not the same leased item as the target leased item.
[0041] The above approach avoids redundant calculations by retaining the matching pairs with the highest correlation strength. Furthermore, when calculating the comprehensive correlation score, the weighted average correlation strength is multiplied by the target attribute feature coverage, allowing the matching quality of highly important attribute features to have a greater impact on the final score. This also avoids the situation where the score is too low due to the sparseness of the target leased item's attributes. In addition, a conflict penalty coefficient is introduced, which automatically reduces the score when attribute features have contradictory attribute values. This achieves a quantitative evaluation through the fusion of multi-dimensional parameters, making the obtained comprehensive correlation score more reliable and thus significantly improving the accuracy of identifying duplicate leased items.
[0042] Furthermore, the preset importance weights of the historical attribute features are obtained in advance through the following steps: S410, pre-set static weights for each historical attribute feature according to the attribute type; wherein, the static weight of historical attribute features of unique identifiers is higher than the static weight of historical attribute features of non-unique identifiers. For example, the serial number is a unique identifier.
[0043] S420, based on the set of historical leased items, counts the frequency of different values of historical attribute corresponding to each historical attribute feature, and calculates the dynamic weight adjustment coefficient of each historical attribute feature based on information entropy.
[0044] Specifically, the formula for calculating the dynamic weight adjustment coefficient is as follows: Where w0 is the dynamic weight adjustment coefficient, N is the total number of historical leased assets, and df j Let L be the frequency of different values of the historical attribute corresponding to the j-th historical attribute feature, and L be the total number of standard attribute features.
[0045] By setting this formula, the smaller the frequency, the larger the numerator. Then, normalization is performed so that the calculated dynamic weight adjustment coefficient can reflect the relative information content of the historical attribute feature in the entire attribute system. The more unique the attribute value, the larger the dynamic weight adjustment coefficient.
[0046] S430, the product of the static weight of each historical attribute feature and the corresponding dynamic weight adjustment coefficient is used as the preset importance weight of the historical attribute feature.
[0047] As described above, by adaptively adjusting the static weights through dynamic weight adjustment coefficients, the bias of expert prior knowledge towards core attribute features is preserved, while information entropy is used to automatically enhance the weights of attribute features with high uniqueness and strong discriminative power, suppressing the influence of commonly repeated attribute features. This allows the final fusion weights to more accurately reflect the actual contribution of each attribute feature in duplicate identification, thereby improving the rationality of the comprehensive association score.
[0048] S5. When the highest comprehensive correlation score is greater than the preset judgment threshold, the target leased item is determined to be the same leased item as the historical leased item with the highest comprehensive correlation score, and the recognition result is output.
[0049] Preferably, the preset judgment threshold meets the following conditions: Where θ is the preset judgment threshold, B is the preset base threshold, α is the preset adjustment range, and F i Let θ be the similarity of the attribute values corresponding to the i-th valid attribute feature pair, δ be a preset high similarity judgment threshold, and I() be an indicator function that takes the value 1 when the attribute value similarity is greater than δ, and 0 otherwise. In specific implementations, the value range of θ is limited to between 0.6 and 0.8.
[0050] As described above, by comparing the highest comprehensive correlation score with the preset judgment threshold, the duplicate identification result can be output adaptively. The preset judgment threshold is limited, and the judgment strictness is adaptively adjusted according to the matching quality of high-weight attribute features. When the proportion of high similarity and high-weight attributes is high, the threshold is reduced to avoid missed judgments; otherwise, the threshold is increased to avoid false judgments. This overcomes the defect that a fixed threshold cannot adapt to different matching quality scenarios and improves the accuracy and reliability of duplicate identification.
[0051] Furthermore, step S5 also includes the following steps: S501: When the highest comprehensive correlation score is not greater than a preset judgment threshold, all target attribute values of the target leased item are concatenated into a target text vector, and all target attribute values of each historical leased item are concatenated into a corresponding historical text vector. For example, the Sentence-BERT model is used to generate the target text vector and the historical text vector.
[0052] S502, calculate the text similarity between the target text vector and each historical text vector. The text similarity is obtained by calculating the cosine similarity.
[0053] S503, if there is text similarity not less than a preset catch-all threshold, the target rental item is marked as a suspected duplicate rental item, and the text similarity not less than the preset catch-all threshold is sorted from largest to smallest, and the historical rental items corresponding to the top k text similarities are selected for review. Those skilled in the art can set the preset catch-all threshold according to actual needs, for example, 0.6.
[0054] S504. If the similarity of all texts is less than the preset fallback threshold, the target leased item is determined to be a newly leased item, and the historical attribute feature association index table is updated.
[0055] As mentioned above, when the highest comprehensive association score is still low, a fallback search is performed using the overall text vector similarity. The semantic information of the complete description of the leased property is used to make up for the missing matching problem caused by missing attribute features or failure of standardization. At the same time, a candidate list is output based on similarity ranking for manual review, and the confirmed new leased properties are dynamically updated to the historical attribute feature association index table, thus realizing the continuous improvement of data.
[0056] In another embodiment, step S5 can also be replaced by the following steps: S51, when the highest comprehensive correlation score is not less than the preset strong judgment threshold, the target leased asset is directly determined to be the same leased asset as the historical leased asset with the corresponding highest comprehensive correlation score. Those skilled in the art can set the preset strong judgment threshold, preset weak judgment threshold, and preset rejection threshold according to actual needs, for example, 0.85, 0.7, and 0.6.
[0057] S52, when the highest comprehensive correlation score is less than the preset strong judgment threshold but not less than the preset weak judgment threshold, the target leased item is marked as a suspected duplicate leased item, and all historical leased items with a comprehensive correlation score not less than the preset weak judgment threshold are screened out for review; wherein, the preset weak judgment threshold is less than the preset strong judgment threshold.
[0058] S53, when the highest comprehensive correlation score is less than the preset weak judgment threshold but not less than the preset denial threshold, the target leased item is marked as a suspected duplicate leased item, and the historical leased item corresponding to the highest comprehensive correlation score is output for review; wherein, the preset denial threshold is less than the preset weak judgment threshold.
[0059] S54. When the highest comprehensive correlation score is less than the preset negation threshold, the target leased item is determined to be a newly leased item.
[0060] The above-mentioned three-level thresholds enable the hierarchical output of duplicate recognition results. When the score is high and unique, it is automatically determined; when the score is medium, the entire candidate list is output for review; when the score is low but not the lowest, the historical rental item with the highest score is output for review; and when the score is extremely low, it is determined to be a newly added rental item. Thus, while ensuring high confidence, it provides differentiated manual review guidance for edge-blurred scenarios, balancing recognition efficiency and accuracy.
[0061] Embodiments of the present invention also provide a non-transitory computer-readable storage medium that can be disposed in an electronic device to store at least one instruction or at least one program related to implementing a method in the method embodiments, wherein the at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided in the above embodiments.
[0062] Embodiments of the present invention also provide an electronic device, including a processor and the aforementioned non-transitory computer-readable storage medium.
[0063] While specific embodiments of the invention have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. It should also be understood that various modifications can be made to the embodiments without departing from the scope and spirit of the invention. The scope of the invention is defined by the appended claims.
Claims
1. A method for identifying duplicate leased items based on feature association, characterized in that, The method includes the following steps: S1. Extract the historical lease item identifier and the historical attribute feature set of each historical lease item from the obtained historical lease text, and construct a historical attribute feature association index table based on the historical attribute feature set; S2, when the user inputs the rental item query text, extract the target rental item identifier and the target attribute feature set of each target rental item from the rental item query text; S3. For each target leased item, use any target attribute feature in the target attribute feature set of the target leased item as the query key, match at least one historical attribute feature from the historical attribute feature association index table, and calculate the association strength between each matched historical attribute feature and the target attribute feature based on the name of the historical attribute feature and the corresponding historical attribute value. S4, aggregate the association strengths corresponding to all target attribute features according to the historical leased items to which the matched historical attribute features belong, and calculate the comprehensive association score between each historical leased item and the target leased item; S5. When the highest comprehensive correlation score is greater than the preset judgment threshold, the target leased item is determined to be the same leased item as the historical leased item with the highest comprehensive correlation score, and the recognition result is output.
2. The method for identifying duplicate leased items based on feature association according to claim 1, characterized in that, In step S1, constructing the historical attribute feature association index table based on the historical attribute feature set includes the following steps: S101, for each historical leased item, standardize each historical attribute feature and its corresponding historical attribute value into a key-value pair format of standard attribute feature name - standardized attribute value; wherein, the standardized attribute value is obtained by processing the historical attribute value according to the attribute type of the historical attribute feature; S102, using the key-value pairs as index keys, establish an inverted index table pointing to the historical leased item identifier as a historical attribute feature association index table.
3. The method for identifying duplicate leased items based on feature association according to claim 1, characterized in that, Step S3 includes the following steps: S301, For any target attribute feature, calculate the semantic similarity between the target attribute feature and each historical attribute feature in the historical attribute feature association index table, and take the historical attribute features with a semantic similarity greater than a preset similarity threshold as the matched historical attribute features. S302, for each matched historical attribute feature, look up the historical attribute value corresponding to the historical attribute feature in the historical attribute feature association index table, and calculate the vector similarity between the found historical attribute value and the target attribute value of the target attribute feature according to the corresponding calculation method based on the attribute type. S303, calculate the association strength between each matched historical attribute feature and the target attribute feature by performing a weighted sum based on the semantic similarity and vector similarity.
4. The method for identifying duplicate leased items based on feature association according to claim 1, characterized in that, Step S4 includes the following steps: S401, for any historical leased property, when multiple target attribute features match the same historical attribute feature of the historical leased property, retain the matching pair between the target attribute feature with the highest correlation strength and the historical attribute feature as the valid attribute feature pair associated with the historical leased property; S402, Obtain all valid attribute feature pairs associated with historical leased assets, forming a set of valid attribute feature pairs corresponding to historical leased assets; S403, Calculate the comprehensive association score between the historical leased assets and the target leased assets based on the set of valid attribute features corresponding to the historical leased assets; wherein, the comprehensive association score meets the following conditions: Where S is the comprehensive correlation score, n is the number of valid attribute feature pairs in the set of valid attribute feature pairs corresponding to historical leased items, and r is the number of valid attribute feature pairs in the set of valid attribute feature pairs. i w represents the correlation strength between the target attribute feature and the historical attribute feature in the i-th valid attribute feature pair. i is the preset importance weight of the historical attribute features in the i-th effective attribute feature pair, c is the target attribute feature coverage corresponding to the effective attribute feature pair set, and p is the preset feature conflict penalty coefficient.
5. The method for identifying duplicate leased items based on feature association according to claim 4, characterized in that, The preset importance weights of the historical attribute features are obtained in advance through the following steps: S410, pre-set static weights for each historical attribute feature according to the attribute type; wherein, the static weight of historical attribute features of unique identifiers is higher than the static weight of historical attribute features of non-unique identifiers. S420, based on the set of historical leased items, counts the frequency of different values of historical attribute corresponding to each historical attribute feature, and calculates the dynamic weight adjustment coefficient of each historical attribute feature based on information entropy. S430, the product of the static weight of each historical attribute feature and the corresponding dynamic weight adjustment coefficient is used as the preset importance weight of the historical attribute feature.
6. The method for identifying duplicate leased items based on feature association according to claim 4, characterized in that, The preset judgment threshold meets the following conditions: Where θ is the preset judgment threshold, B is the preset base threshold, α is the preset adjustment range, and F i Let δ be the similarity of the attribute values corresponding to the i-th valid attribute feature pair, δ be a preset high similarity judgment threshold, and I() be an indicator function. When the similarity of the attribute values is greater than δ, the value is 1; otherwise, the value is 0.
7. The method for identifying duplicate leased items based on feature association according to claim 1, characterized in that, Step S5 also includes the following steps: S501, when the highest comprehensive correlation score is not greater than the preset judgment threshold, all target attribute values of the target leased item are concatenated into a target text vector, and all target attribute values of each historical leased item are concatenated into the corresponding historical text vector. S502, calculate the text similarity between the target text vector and each historical text vector; S503, if there is a text similarity of not less than the preset catch-all threshold, mark the target rental item as a suspected duplicate rental item, and sort the text similarity of not less than the preset catch-all threshold from largest to smallest, and select the historical rental items corresponding to the top k text similarities for review. S504. If the similarity of all texts is less than the preset fallback threshold, the target leased item is determined to be a newly leased item, and the historical attribute feature association index table is updated.
8. The method for identifying duplicate leased items based on feature association according to claim 1, characterized in that, Step S5 can also be replaced by the following steps: S51, when the highest comprehensive correlation score is not less than the preset strong judgment threshold, the target leased item and the historical leased item with the corresponding highest comprehensive correlation score are directly judged to be the same leased item; S52, when the highest comprehensive correlation score is less than the preset strong judgment threshold but not less than the preset weak judgment threshold, the target leased item is marked as a suspected duplicate leased item, and all historical leased items with a comprehensive correlation score not less than the preset weak judgment threshold are filtered out for review; where the preset weak judgment threshold is less than the preset strong judgment threshold. S53, when the highest comprehensive correlation score is less than the preset weak judgment threshold but not less than the preset rejection threshold, the target leased item is marked as a suspected duplicate leased item, and the historical leased item corresponding to the highest comprehensive correlation score is output for review; wherein, the preset rejection threshold is less than the preset weak judgment threshold. S54. When the highest comprehensive correlation score is less than the preset negation threshold, the target leased item is determined to be a newly leased item.
9. A non-transitory computer-readable storage medium, wherein the storage medium stores at least one instruction or at least one program segment, characterized in that, The at least one instruction or the at least one program segment is loaded and executed by the processor to implement the feature association-based duplicate lease identification method as described in any one of claims 1-8.
10. An electronic device, characterized in that, Includes a processor and the non-transitory computer-readable storage medium as described in claim 9.