Contract management method and equipment based on information retrieval and medium
Through deep feature extraction and robust collaborative matrix decomposition, the problem of large-scale computing overhead in similar contract scenarios is solved, the efficiency and accuracy of contract management are improved, and efficient contract information retrieval and management are achieved.
Patent Information
- Application Number
- CN202411032994.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-11-15
AI Technical Summary
In the scenario of large-scale similar contracts, the information retrieval of traditional contract data consumes a huge calculation overhead, resulting in the identification efficiency of similar contracts that cannot meet the needs of a large number of contracts, and the efficiency of contract management needs to be improved.
Using a contract management method based on information retrieval, the multimodal contract same sample data is obtained, deep feature extraction and robust collaborative matrix decomposition are performed, the objective function and information retrieval index are determined, and the pre-constructed multimodal contract same sample library is used for similarity. Match, generate a similarity sequence for management operations.
It improves the accuracy and efficiency of information retrieval, reduces manual intervention, realizes efficient contract management, can adapt to the processing needs of large-scale contract databases, and improves user experience and personalized services for contract management.
Smart Images

Figure CN118964443A_ABST
Abstract
Description
Technical Field
[0001] The present specification relates to the technical field of contract management, and in particular to a contract management method, device and medium based on information retrieval. Background Art
[0002] With the development of the economy and the progress of society, contracts have become an indispensable legal text in business activities, and the number of contract signings has also increased exponentially in recent years. Retrieving similar contracts from a large number of contracts through manual screening requires huge human, material and other resources, resulting in the inability of relevant practitioners to fully grasp and utilize existing contract data, which to a certain extent affects the efficient advancement of related businesses. Contract information retrieval is an indispensable part of smart contract management. Similar contracts can be identified through intra-modal and cross-modal retrieval of multimodal contract data. The identification of similar contracts helps to quickly classify and archive, allowing contract managers to handle a large number of contracts more efficiently.
[0003] For similarity retrieval of contract information, traditional retrieval technology can only discover shallow semantic relationships and cannot obtain complex semantic association information between contract images and contract text data. In addition, in order to maintain accuracy, traditional retrieval methods usually convert multimodal information into real-valued representations, and then use calculation formulas such as cosine distance to measure the similarity between multimodal data to complete the similar contract retrieval task. It is worth noting that when the scale of similar contract data to be retrieved is large, traversing the contract database will consume a huge computational overhead, so it is not suitable for large-scale similar contract retrieval scenarios. In recent years, with the development of machine learning, hash learning based on nearest neighbor search has been widely used in multimedia retrieval matching due to its advantages such as low storage cost and high computational efficiency. Existing unsupervised hashing methods focus more on the learning stage of hash coding and ignore the feature extraction stage, which is crucial for the retrieval of similar contracts.
[0004] Therefore, in the scenario of large-scale similar contracts, information retrieval of traditional contract data consumes huge computing overhead, and the efficiency of identifying similar contracts cannot meet the scenario requirements of a large number of contracts, resulting in the need to improve contract management efficiency. Summary of the invention
[0005] One or more embodiments of the present specification provide a contract management method, device and medium based on information retrieval, which are used to solve the following technical problems: in the scenario of large-scale similar contracts, the information retrieval of traditional contract data consumes a huge computing overhead, resulting in the recognition efficiency of similar contracts being unable to meet the scenario requirements of a large number of contracts, and the contract management efficiency needs to be improved.
[0006] One or more embodiments of this specification adopt the following technical solutions:
[0007] One or more embodiments of the present specification provide a contract management method based on information retrieval, the method comprising: obtaining multimodal contract sample data corresponding to a contract to be managed, wherein the multimodal contract sample data comprises any one or more of contract image data and contract text data; performing deep feature extraction on the multimodal contract sample data, and determining an objective function corresponding to the multimodal contract sample data through robust collaborative matrix decomposition; determining an information retrieval index corresponding to the multimodal contract sample data based on the objective function, and performing contract information similarity matching in a pre-constructed multimodal contract sample library through the information retrieval index to determine a similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library; and performing management operations on the contract to be managed based on the similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library.
[0008] Furthermore, deep feature extraction is performed on the multimodal contract sample data, and the objective function corresponding to the multimodal contract sample data is determined through robust collaborative matrix decomposition, specifically including: performing deep feature extraction on the multimodal contract sample data through an image encoder and / or a text encoder to generate deep image features and / or deep text features; and determining the objective function corresponding to the multimodal contract sample data through robust collaborative matrix decomposition based on the deep image features and / or deep text features.
[0009] Furthermore, the objective function corresponding to the multimodal contract sample data is determined through robust collaborative matrix decomposition, which specifically includes: performing robust collaborative matrix decomposition on the multimodal contract sample data to learn the real-valued public representation of the multimodal contract sample data; based on the real-valued public representation, introducing orthogonal matrices and asymmetric strategies to determine the objective function corresponding to the multimodal contract sample data.
[0010] Furthermore, based on the real-valued public representation, an orthogonal matrix and an asymmetric strategy are introduced to determine the objective function corresponding to the multimodal contract sample data, specifically including: by introducing an orthogonal matrix, the quantization error between the real-valued public representation and the hash code is minimized to transfer the semantic information learned in the robust common subspace to the Hamming space; according to the deep image features and / or deep text features, intra-modal and inter-modal similarity matrices are constructed, and the similarity matrices are subjected to second-order similarity to determine the inter-modal joint similarity matrix; an asymmetric strategy is adopted to embed the inter-modal joint similarity matrix into the learning process of the hash code to transfer the intra-modal semantic information and inter-modal complementary information extracted by the model to the Hamming space to generate a hash code.
[0011] Furthermore, based on the objective function, an information retrieval index corresponding to the multimodal contract sample data is determined, specifically including: determining a corresponding hash function according to the objective function, projecting the multimodal contract sample data based on the hash function, and generating a corresponding retrieval hash code.
[0012] Furthermore, in a pre-constructed multimodal contract sample library, contract information similarity matching is performed through the information retrieval index to determine a similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library, specifically including: determining the information retrieval index, and obtaining sample information retrieval indicators corresponding to multiple existing contracts in the multimodal contract sample library; determining a similarity indicator between the information retrieval indicator and each of the sample information retrieval indicators based on the information retrieval indicator and multiple sample information retrieval indicators; and sorting the multiple existing contracts according to the similarity indicator to determine the similarity sequence.
[0013] Furthermore, when the information retrieval indicator is a retrieval hash code, the similarity indicator between the information retrieval indicator and each of the sample information retrieval indicators is determined based on the information retrieval indicator and multiple sample information retrieval indicators, specifically including: calculating the Hamming distance between the retrieval hash code and the sample hash code corresponding to each of the existing contracts, and obtaining multiple Hamming distances to determine the similarity indicator between the information retrieval indicator and each of the sample information retrieval indicators.
[0014] Further, according to the similarity sequence between the contract to be managed and a plurality of existing contracts in the multimodal contract sample library, a management operation is performed on the contract to be managed, specifically including: determining the management operation corresponding to the contract to be managed according to the similarity sequence between the contract to be managed and a plurality of existing contracts in the multimodal contract sample library and a preset similarity threshold, wherein the management operation includes a contract storage operation and a contract update operation; when there is at least one specified similarity in the similarity sequence greater than the preset similarity threshold, determining similar sample modal data of similar existing contracts corresponding to the specified similarity, wherein the sample modal data includes image samples and text samples; any one of the samples; determining the modal information of the multimodal contract sample data, wherein the modal information includes any one of an image and a text; when there is a similar existing contract with the same modality as the modal information of the multimodal contract sample data in the similar sample modal data, determining that the management operation corresponding to the contract to be managed is a contract update operation; when there is no similar existing contract with the same modality as the modal information of the multimodal contract sample data in the similar sample modal data, or, when the similarities in the similarity sequence are not greater than a preset similarity threshold, determining that the management operation corresponding to the contract to be managed is a contract warehousing operation.
[0015] One or more embodiments of this specification provide a contract management device based on information retrieval, including:
[0016] at least one processor; and,
[0017] a memory communicatively connected to the at least one processor; wherein,
[0018] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the above method.
[0019] One or more embodiments of the present specification provide a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to execute the above method.
[0020] At least one of the above technical solutions adopted in the embodiments of this specification can achieve the following beneficial effects: through the above technical solution, by simultaneously considering the contract image data and the contract text data (which may even include data of more modes), the system can capture richer and more comprehensive contract information, thereby improving the accuracy of information retrieval; using deep learning technology to extract deep features of multimodal data, these features can better represent the essence of the data than traditional manual features, and help to more accurately represent the content of the contract; in the training stage, by integrating the collaborative robust matrix decomposition model, asymmetric strategy, fusion similarity matrix and low-rank constraints, a common semantic representation and double hash function are obtained, and the training is discretely optimized. The whole process is highly automated and intelligent, which reduces manual intervention and improves the efficiency and accuracy of contract management. Management operation decisions based on similarity sequences are more accurate, avoiding unnecessary duplication of work while ensuring timely updating and storage of contract information. Due to the use of efficient feature extraction and similarity matching algorithms, the system can quickly respond to users' contract management requests and improve user experience. By considering the contract management needs of different users, the system can provide more personalized services to meet the contract management requirements in different scenarios. It has good scalability and can adapt to the processing needs of large-scale contract sample libraries, effectively improving the efficiency of contract management. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:
[0022] Figure 1 A flowchart of a contract management method based on information retrieval provided in an embodiment of this specification;
[0023] Figure 2 A schematic diagram of the structure of a contract management device based on information retrieval provided in an embodiment of this specification. DETAILED DESCRIPTION
[0024] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.
[0025] The embodiment of this specification provides a contract management method based on information retrieval. It should be noted that the execution subject in the embodiment of this specification can be a server or any device with data processing capabilities. Figure 1 A flowchart of a contract management method based on information retrieval is provided in the embodiment of this specification, such as Figure 1 As shown, it mainly includes the following steps:
[0026] Step S101, obtaining multimodal contract sample data corresponding to the contract to be managed.
[0027] The multimodal contract sample data includes any one or more of contract image data and contract text data;
[0028] In one embodiment of the present specification, multimodal contract sample data corresponding to the contract to be managed is obtained. The multimodal contract sample data includes any one or more of contract image data and contract text data. Multimodal data refers to data obtained from different fields or perspectives for the same description object, where each data form or source can be regarded as a modality.
[0029] Contract image data can be obtained by scanning paper contracts, taking photos, or extracting from electronic documents. Use a high-resolution scanner to scan paper contracts into electronic image files, ensuring that the image is clear, without blur, distortion or missing parts during the scanning process. The scanned image file should be saved in a common image format, such as JPEG, PNG or TIFF, for subsequent processing and storage. In a well-lit environment, use a mobile phone or a professional camera to take a photo of the paper contract. Ensure that the photo is clear and the contract content is complete and unobstructed. Similarly, after taking the photo, the photo should be saved in a suitable image format, and some image preprocessing (such as cropping, rotation, denoising, etc.) may be required to improve the image quality. If the contract already exists in the form of an electronic document (such as PDF, Word, etc.), OCR (optical character recognition) technology can be used to extract image data from the document, which usually involves converting specific pages or areas in the document into image format. Contract text data acquisition can directly extract text from electronic documents. For contracts that are already electronic documents, you can directly copy and paste the text content from the document, or use text extraction tools to obtain data in plain text format. You can also use OCR technology to recognize text in images. For contract image data, you can use OCR technology to recognize the text in the image and convert it into an editable text format. OCR technology can handle multiple languages and complex layouts, improving the accuracy and efficiency of text recognition. In some cases, it may be necessary to manually enter the contract text data, especially when the contract image quality is poor or the OCR recognition effect is poor. Although manual entry is time-consuming and labor-intensive, it can ensure the accuracy and completeness of the data.
[0030] Step S102, performing deep feature extraction on the multimodal contract sample data, and determining the objective function corresponding to the multimodal contract sample data through robust collaborative matrix decomposition;
[0031] Performing deep feature extraction on the multimodal contract sample data and determining the objective function corresponding to the multimodal contract sample data through robust collaborative matrix decomposition, specifically comprising: performing deep feature extraction on the multimodal contract sample data through an image encoder and / or a text encoder to generate deep image features and / or deep text features; and determining the objective function corresponding to the multimodal contract sample data through robust collaborative matrix decomposition based on the deep image features and / or deep text features.
[0032] The objective function corresponding to the multimodal contract sample data is determined through robust collaborative matrix decomposition, specifically including: performing robust collaborative matrix decomposition on the multimodal contract sample data to learn the real-valued public representation of the multimodal contract sample data; based on the real-valued public representation, introducing orthogonal matrices and asymmetric strategies to determine the objective function corresponding to the multimodal contract sample data.
[0033] Based on the real-valued public representation, an orthogonal matrix and an asymmetric strategy are introduced to determine the objective function corresponding to the multimodal contract sample data, specifically including: minimizing the quantization error between the real-valued public representation and the hash code by introducing an orthogonal matrix to transfer the semantic information learned in the robust common subspace to the Hamming space; constructing intra-modal and inter-modal similarity matrices based on the deep image features and / or deep text features, and performing second-order similarity on the similarity matrices to determine the inter-modal joint similarity matrix; adopting an asymmetric strategy to embed the inter-modal joint similarity matrix into the learning process of the hash code to transfer the intra-modal semantic information and inter-modal complementary information extracted by the model to the Hamming space to generate a hash code.
[0034] In one embodiment of the present specification, the contract image data and contract text instance data to be retrieved are obtained. Assume that the contract image and text data sets are A single data point is in, is the contract image instance data, It is the contract text instance data.
[0035] Furthermore, the image encoder based on the model CLIP generates deep image features of the contract to be retrieved, and the text encoder based on the pre-trained model CLIP generates deep text features: in, denote the image and text encoders of the CLIP model, respectively. and Represent the extracted contract image and text features respectively.
[0036] The objective function consists of five items. The first item learns a low-dimensional common representation through robust collaborative matrix decomposition; the second item adopts an asymmetric strategy to transfer the similarity information of fused semantics to the Hamming space; the third item minimizes the quantization error between the common representation and the hash code, and imposes orthogonal constraints on the projection matrix; the fourth item learns two projection matrices, namely, modality-specific and modality-consistent hash functions, and imposes smoothing and low-rank constraints on them respectively; the fifth item is the regularization term to avoid overfitting problems.
[0037] In the objective function, the goal of robust collaborative matrix decomposition is to explore a common subspace of multimodal data and learn a common representation of heterogeneous contract samples. In real-world contract image or text datasets, noise and outliers are inevitable, and the loss function used by previous cross-modal hashing methods based on matrix decomposition, that is, the minimum square error function, is highly sensitive to noisy data and outliers. Therefore, the objective function introduces l 2,p norm as the loss function to learn a real-valued common representation of multimodal contract data with robustness: Among them, U1∈R d×k is the basis matrix of the image modality, U2∈R d×k is the basis matrix of the text modality, V∈R k×n is the public representation matrix.
[0038] In the objective function, by introducing an orthogonal matrix to minimize the quantization error between the real-valued public representation and the hash code, the semantic information learned in the robust common subspace is better transferred to the Hamming space, maintaining the discrete properties of the hash code. This process can be described as follows: Among them, B is the unified hash code of the training samples, R∈R k×k is an orthogonal matrix.
[0039] Furthermore, in order to reduce the cost of manual annotation and improve training efficiency, the algorithm uses the deep features of contract images and texts containing rich semantic information extracted by the pre-trained model CLIP to construct the intra-modal and inter-modal similarity matrix, which is used to describe the neighborhood structure information of contract images and texts and the deep semantic associations across modalities. The similarity matrix within the contract image modality is expressed as The similarity matrix of text modality is in, Defined as:
[0040]
[0041] Furthermore, in order to maintain the semantic relationship between different modalities and achieve mutual complementation, the method further performs second-order similarity on the modality-specific similarity matrix and fuses it into an inter-modality joint similarity matrix:
[0042]
[0043] In the objective function, an asymmetric strategy is used to embed the fused similarity matrix into the learning process of hash coding, so that the intra-modal semantic information extracted by the pre-trained model and the inter-modal complementary information are transferred to the Hamming space to generate a more discriminative hash code:
[0044] Here, k is the length of the hash code.
[0045] In the objective function, learning the modality-specific hash function can obtain the intra-modality information, and learning the modality-consistent hash function can preserve the diversity and complementary information of each modality data of the contract in the hash encoding:
[0046] Among them, P irepresents the modality-specific mapping matrix, and Q represents the inter-modality consistent mapping matrix. For the modality-specific mapping matrix, the Frobenius norm is used to constrain it to prevent overfitting during the training process; for the consistent mapping matrix, the nuclear norm is used to constrain it to a low rank so that it retains more common information; α, β, μ, and θ are hyperparameters
[0047] The final objective function is:
[0048]
[0049] stB∈{-1,1} k×n ,RR T =I
[0050] Step S103, based on the objective function, determine the information retrieval index corresponding to the multimodal contract sample data, and in the pre-constructed multimodal contract sample library, perform contract information similarity matching through the information retrieval index to determine the similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library.
[0051] Based on the objective function, an information retrieval index corresponding to the multimodal contract sample data is determined, specifically including: determining a corresponding hash function according to the objective function, projecting the multimodal contract sample data based on the hash function, and generating a corresponding retrieval hash code.
[0052] In one embodiment of the present specification, the corresponding hash function is determined according to the objective function, and the deep image features corresponding to the contract image data and / or the deep text features corresponding to the contract text data are linearly transformed according to the objective function. The hash code corresponding to the contract image data is generated by right-multiplying the deep image features corresponding to the contract image data to be retrieved by P1 and Q and then taking the sign function sign to change it to 1 and -1, or the deep text features corresponding to the contract text data to be retrieved are right-multiplied by P2 and Q and then taking the sign function sign to generate the hash code of the query contract text data. At the same time, the multimodal existing contract data corresponding to the existing contracts in the multimodal contract sample library are projected according to the above process to determine the retrieval hash code corresponding to each multimodal existing contract data in the multimodal contract sample library.
[0053] In a pre-constructed multimodal contract sample library, the information retrieval index is used to perform contract information similarity matching to determine a similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library, specifically including: determining the information retrieval index, and obtaining sample information retrieval indicators corresponding to multiple existing contracts in the multimodal contract sample library; determining a similarity indicator between the information retrieval indicator and each of the sample information retrieval indicators based on the information retrieval index and multiple sample information retrieval indicators; and sorting the multiple existing contracts according to the similarity index to determine the similarity sequence.
[0054] When the information retrieval indicator is a retrieval hash code, the similarity indicator between the information retrieval indicator and each of the sample information retrieval indicators is determined based on the information retrieval indicator and multiple sample information retrieval indicators, specifically including: calculating the Hamming distance between the retrieval hash code and each sample hash code corresponding to the existing contract, and obtaining multiple Hamming distances to determine the similarity indicator between the information retrieval indicator and each of the sample information retrieval indicators.
[0055] Feature vectors are extracted for the contract image data and text data respectively. For fast retrieval, the high-dimensional feature vectors are converted into low-dimensional hash codes. Retrieval indicators, such as the Hamming distance between hash codes, are defined to measure the similarity between the contract to be retrieved and the existing contracts. For each existing contract in the multimodal contract sample library, its image and text features are extracted and the corresponding hash code is generated. These hash codes are used as sample information retrieval indicators to compare with the hash code of the contract to be retrieved. The Hamming distance between the hash code of the contract to be retrieved and the hash code of each existing contract is calculated. The smaller the Hamming distance, the more similar the two hash codes are, that is, the higher the similarity between the contract to be retrieved and the existing contracts. The existing contracts are sorted according to the Hamming distance to obtain a similarity sequence. The first existing contract in the sequence has the highest similarity with the contract to be retrieved, and so on.
[0056] Through the above technical scheme, in the training stage, by integrating the collaborative robust matrix decomposition model, asymmetric strategy, fusion similarity matrix and low-rank constraint, the public semantic representation and double hash function are obtained, and the hash code of the training data is discretely optimized; in the query stage, the double hash function obtained by the objective function is used to generate the hash code of the query sample to realize cross-modal retrieval; by converting the high-dimensional feature vector into a low-dimensional hash code, the computational complexity of the retrieval is significantly reduced, thereby accelerating the retrieval speed; the Hamming distance calculation between hash codes is simple and fast, making it possible to perform fast retrieval in a large-scale contract sample library; the comprehensive use of information in the two modes of contract image and text can more comprehensively reflect the characteristics and content of the contract, thereby improving the accuracy of retrieval; compared with the single-modality retrieval method, multimodal retrieval can capture more contract similarities and reduce missed detection and false detection; the generation of similarity sequences enables contract managers to clearly see the similarity relationship between the contract to be managed and the existing contract, so as to make more reasonable decisions on warehousing or updating; through the automated and intelligent contract management process, manual intervention and subjective judgment are reduced, and the objectivity and fairness of contract management are improved.
[0057] Step S104: performing management operations on the contract to be managed according to the similarity sequence between the contract to be managed and a plurality of existing contracts in the multimodal contract sample library.
[0058] According to the similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library, a management operation is performed on the contract to be managed, specifically including: according to the similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library and a preset similarity threshold, determining the management operation corresponding to the contract to be managed, wherein the management operation includes a contract storage operation and a contract update operation; when there is at least one specified similarity in the similarity sequence greater than the preset similarity threshold, determining similar sample modal data of similar existing contracts corresponding to the specified similarity, wherein the sample modal data includes image samples and text samples; any one of the following; determining the modal information of the multimodal contract sample data, wherein the modal information includes any one of an image and a text; when there is a similar existing contract with the same modality as the modal information of the multimodal contract sample data in the similar sample modal data, determining that the management operation corresponding to the contract to be managed is a contract update operation; when there is no similar existing contract with the same modality as the modal information of the multimodal contract sample data in the similar sample modal data, or when the similarities in the similarity sequence are not greater than a preset similarity threshold, determining that the management operation corresponding to the contract to be managed is a contract warehousing operation.
[0059] In one embodiment of the present specification, the similarity between the contract to be managed and multiple existing contracts in the multimodal contract sample library is calculated. According to the similarity, a similarity sequence is generated, and the sequence is arranged from high to low according to the similarity. A preset similarity threshold is set to determine whether the similarity between the contract to be managed and the existing contract is high enough to perform an update operation. The similarity sequence is traversed to check whether there is at least one specified similarity greater than the preset similarity threshold. When there is a specified similarity greater than the preset similarity threshold, the similar existing contract corresponding to the similarity is determined. Sample modal data of these similar existing contracts are extracted, including any one of image samples and text samples. The modal information contained in the multimodal contract sample data of the contract to be managed is determined, that is, any one of the image and text. Check whether there is a similar existing contract in the similar sample modal data with the same modality as the multimodal contract sample data of the contract to be managed. If so, it means that the contract to be managed has a sufficiently high similarity with an existing contract in the same modality, so the management operation is determined to be a contract update operation. If it does not exist, or all similarities in the similarity sequence are not greater than the preset similarity threshold, it means that the content of the contract to be managed is quite different from that of the existing contract, or there is no contract data under this mode in the existing multimodal contract sample library, so the management operation is determined to be a contract storage operation. In addition to the above management process, other management operations such as classifying and archiving the contract to be managed in the existing multimodal contract sample library can also be performed according to the similarity sequence.
[0060] Through the above technical scheme, by considering both the contract image data and the contract text data (may even include data of more modalities), the system can capture richer and more comprehensive contract information, thereby improving the accuracy of information retrieval; using deep learning technology to extract deep features of multimodal data, these features can better represent the essence of data than traditional manual features, and help to more accurately represent the content of the contract; in the training stage, by integrating the collaborative robust matrix decomposition model, asymmetric strategy, fusion similarity matrix and low-rank constraints, a common semantic representation and double hash function are obtained to discretely optimize the hash encoding of the training data; the entire process is highly automated and intelligent, reducing manual intervention and improving the efficiency and accuracy of contract management; management operation decisions based on similarity sequences are more accurate, avoiding unnecessary duplication of work, and ensuring the timely update and storage of contract information; due to the use of efficient feature extraction and similarity matching algorithms, the system can quickly respond to users' contract management requests and improve user experience; by considering the contract management needs of different users, the system can provide more personalized services to meet the contract management requirements in different scenarios; it has good scalability and can adapt to the processing needs of large-scale contract sample libraries, effectively improving the efficiency of contract management.
[0061] The embodiment of this specification also provides a contract management device based on information retrieval, such as Figure 2 As shown, the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above method.
[0062] The embodiments of the present specification also provide a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to execute the above method.
[0063] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0064] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0065] The devices and media provided in the embodiments of this specification correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.
[0066] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
[0067] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0068] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0069] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0070] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0071] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0072] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0073] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0074] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.
Claims
1. A contract management method based on information retrieval, characterized in that: The method comprises: Acquire multimodal contract sample data corresponding to the contract to be managed, wherein the multimodal contract sample data includes any one or more of contract image data and contract text data; Performing deep feature extraction on the multimodal contract sample data, and determining the objective function corresponding to the multimodal contract sample data through robust collaborative matrix decomposition; Based on the objective function, determining the information retrieval index corresponding to the multimodal contract sample data, and in a pre-constructed multimodal contract sample library, performing contract information similarity matching by using the information retrieval index to determine a similarity sequence between the contract to be managed and a plurality of existing contracts in the multimodal contract sample library; According to the similarity sequence between the contract to be managed and a plurality of existing contracts in the multimodal contract sample library, a management operation is performed on the contract to be managed.
2. A contract management method based on information retrieval according to claim 1, characterized in that: Performing deep feature extraction on the multimodal contract sample data and determining the objective function corresponding to the multimodal contract sample data through robust collaborative matrix decomposition specifically includes: Performing deep feature extraction on the multimodal contract sample data through an image encoder and / or a text encoder to generate deep image features and / or deep text features; According to the deep image features and / or deep text features, the objective function corresponding to the multimodal contract sample data is determined through robust collaborative matrix decomposition.
3. The information retrieval-based contract management method according to claim 1, characterized in that: By means of robust collaborative matrix decomposition, the objective function corresponding to the multimodal contract sample data is determined, specifically including: Performing robust collaborative matrix factorization on the multimodal contract sample data to learn a real-valued common representation of the multimodal contract sample data; Based on the real-valued public representation, an orthogonal matrix and an asymmetric strategy are introduced to determine the objective function corresponding to the multimodal contract sample data.
4. A contract management method based on information retrieval according to claim 3, characterized in that: Based on the real-valued public representation, an orthogonal matrix and an asymmetric strategy are introduced to determine the objective function corresponding to the multimodal contract sample data, specifically including: By introducing an orthogonal matrix, the quantization error between the real-valued public representation and the hash code is minimized to transfer the semantic information learned in the robust public subspace to the Hamming space; Constructing intra-modality and inter-modality similarity matrices based on the deep image features and / or the deep text features, and performing second-order similarity on the similarity matrices to determine an inter-modality joint similarity matrix; An asymmetric strategy is adopted to embed the inter-modal joint similarity matrix into the learning process of hash coding, so as to transfer the intra-modal semantic information and inter-modal complementary information extracted by the model to the Hamming space to generate hash coding.
5. The information retrieval-based contract management method according to claim 1, characterized in that: Based on the objective function, determining the information retrieval index corresponding to the multimodal contract sample data specifically includes: A corresponding hash function is determined according to the objective function, and the multimodal contract sample data is projected based on the hash function to generate a corresponding retrieval hash code.
6. A contract management method based on information retrieval according to claim 1, characterized in that: In the pre-built multimodal contract sample library, the contract information similarity matching is performed by using the information retrieval index to determine the similarity sequence between the contract to be managed and multiple existing contracts in the multimodal contract sample library, specifically including: Determine the information retrieval index, and obtain sample information retrieval indexes corresponding to a plurality of existing contracts in the multimodal contract sample library; Determining a similarity index between the information retrieval index and each of the sample information retrieval indexes according to the information retrieval index and the plurality of sample information retrieval indexes; The multiple existing contracts are sorted according to the similarity index to determine the similarity sequence.
7. A contract management method based on information retrieval according to claim 6, characterized in that: When the information retrieval indicator is a retrieval hash code, determining a similarity indicator between the information retrieval indicator and each of the sample information retrieval indicators according to the information retrieval indicator and the plurality of sample information retrieval indicators specifically includes: The Hamming distance between the retrieval hash code and the sample hash code corresponding to each of the existing contracts is calculated to obtain a plurality of the Hamming distances to determine a similarity index between the information retrieval index and each of the sample information retrieval indexes.
8. A contract management method based on information retrieval according to claim 1, characterized in that: According to the similarity sequence between the contract to be managed and a plurality of existing contracts in the multimodal contract sample library, a management operation is performed on the contract to be managed, specifically including: Determine the management operation corresponding to the contract to be managed according to the similarity sequence and the preset similarity threshold between the contract to be managed and a plurality of existing contracts in the multimodal contract sample library, wherein the management operation includes a contract storage operation and a contract update operation; When there is at least one specified similarity in the similarity sequence that is greater than a preset similarity threshold, determining similar sample modal data of similar existing contracts corresponding to the specified similarity, wherein the sample modal data includes any one of an image sample and a text sample; Determining modal information of the multimodal contract sample data, wherein the modal information includes any one of an image and a text; When there is a similar existing contract in the similar sample modality data with the modality information of the multimodal contract sample data in the same modality, determining that the management operation corresponding to the contract to be managed is a contract update operation; When there is no similar existing contract with the same modality as the modal information of the multimodal contract sample data in the similar sample modal data, or when the similarities in the similarity sequence are not greater than a preset similarity threshold, it is determined that the management operation corresponding to the contract to be managed is a contract warehousing operation.
9. A contract management device based on information retrieval, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the method according to any one of claims 1 to 8.
10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured to execute the method according to any one of claims 1 to 8.
Citation Information
Patent Citations
Mapping dictionary learning based cross-modality hashing search method
CN107256271A
Discrete Hash retrieval method based on robust matrix factorization
CN113407661A
Cross-modal retrieval method based on similarity zero sample hash
CN114943017A
Data retrieval method, system and equipment based on triple matrix decomposition
CN115374165A
Contract-based automatic rule matching algorithm
CN116955536A
Cited By
Intelligent processing method and system for judicial batch filing
CN120317230A
Intelligent contract management method and device for multi-modal large model, and medium
CN120523784A
