A method, device, computing device and storage medium for revising clauses

Abstract

The present application discloses a method for revising clauses, including: obtaining the clauses to be reviewed, and determining at least one clause revision record that matches the clauses to be reviewed in the clause revision library, wherein the clause revision library includes multiple clause revision records, and each clause revision record includes the original clause and the revised content for the original clause; then, using the at least one clause revision record determined, the clause to be reviewed is revised. In this way, for each clause that the user needs to review, the automatic review and revision of the clause can be achieved, which can not only effectively reduce the labor cost, but also avoid the review results of the clause from being affected by human factors, so as to protect the interests of the user as much as possible. In addition, the present application also provides corresponding devices, computing devices and storage media.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, computing device and storage medium for revising clauses. Background Art

[0002] In application scenarios such as enterprises, institutions, and financial institutions, there is usually a need to review the terms in a large number of documents. For example, after Enterprise A and Enterprise B draft a commercial contract, both parties usually need to review the various terms recorded in the contract to avoid losses caused by unreasonable content of some terms in the contract.

[0003] At present, manual review is usually used to proofread and revise the clauses in the document one by one. However, each document usually includes a large number of clauses, and in some scenarios, a large number of documents need to be manually reviewed. This not only makes the workload of manual review large and the labor cost high, but also the effect of manual review of clauses is easily affected by human factors, such as human review omissions, which can easily cause serious loss of interests to users. Summary of the invention

[0004] The embodiments of the present application provide a method, apparatus, computing device and storage medium for revising clauses to realize automatic review of revised clauses, thereby effectively reducing labor costs and improving the quality of clause review.

[0005] In a first aspect, an embodiment of the present application provides a method for revising a clause, the method comprising:

[0006] Get the terms to be reviewed;

[0007] Determine at least one clause revision record in a clause revision library that matches the clause to be reviewed, wherein the clause revision library includes multiple clause revision records, and each clause revision record includes an original clause and revised content for the original clause;

[0008] The at least one clause revision record is used to revise the clause to be reviewed.

[0009] In a possible implementation manner, determining at least one clause revision record in the clause revision library that matches the clause to be reviewed includes:

[0010] Determine at least one clause revision record in the clause revision library whose similarity with the clause to be reviewed meets a preset condition.

[0011] In a possible implementation manner, determining at least one clause revision record in the clause revision library whose similarity with the clause to be reviewed meets a preset condition includes:

[0012] respectively calculating the similarity between the clause to be reviewed and the original clauses included in each clause revision record belonging to the target semantic category in the clause revision library, wherein the semantics of the clause to be reviewed belongs to the target semantic category;

[0013] Based on the similarity between the clause to be reviewed and the original clauses included in each clause revision record, at least one clause revision record is determined from multiple clause revision records belonging to the target semantic category, and the similarity between the original clauses included in each clause revision record in the at least one clause revision record and the clause to be reviewed is higher than the similarity between the original clauses included in other clause revision records in the clause revision library and the clause to be reviewed.

[0014] In a possible implementation, the clause revision library includes clause revision records of multiple semantic categories, the multiple semantic categories include the target category, and the method further includes:

[0015] Acquire historical revision data, wherein the historical revision data includes revision records of clauses revised for multiple original clauses within a historical time period;

[0016] Vectorizing the historical revision data to obtain a vectorized representation corresponding to the historical revision data;

[0017] According to the vectorized representation corresponding to the historical revision data, a plurality of clause revision records included in the historical revision data are clustered to determine the semantic category to which each clause revision record in the historical revision data belongs.

[0018] In a possible implementation manner, the step of revising the to-be-reviewed clause by using the at least one clause revision record includes:

[0019] Presenting a recommendation interface, wherein the recommendation interface includes the at least one clause revision record;

[0020] In response to a user's selection operation on the recommendation interface for the clause revision record, determining a target clause revision record from the at least one clause revision record;

[0021] The terms to be reviewed shall be revised according to the revision contents in the revision record of the target terms.

[0022] In a possible implementation manner, the step of revising the to-be-reviewed clause by using the at least one clause revision record includes:

[0023] Determine a target clause revision record from the at least one clause revision record, wherein the original clause in the target clause revision record has the greatest similarity with the clause to be reviewed;

[0024] The terms to be reviewed shall be revised according to the revision contents in the revision record of the target terms.

[0025] In a possible implementation manner, before revising the clause to be reviewed, the method further includes:

[0026] Matching the to-be-reviewed clauses with the trust clauses in the trust clause library;

[0027] Then, the step of determining at least one clause revision record in the clause revision library that matches the clause to be reviewed includes:

[0028] When there is no trust clause matching the clause to be reviewed in the trust clause library, at least one clause revision record matching the clause to be reviewed in the clause revision library is determined.

[0029] In a possible implementation, the method further includes:

[0030] Acquire a collection clause library, wherein the collection clause library includes a plurality of candidate clauses;

[0031] Clustering the multiple candidate clauses according to the semantics of the multiple candidate clauses to obtain multiple clusters, wherein the semantics of the candidate clauses in each cluster are similar, and the semantics of the candidate clauses in different clusters are different;

[0032] Candidate clauses satisfying the trust condition in each cluster are determined as trust clauses, and the trust clauses are added to the trust clause library.

[0033] In a second aspect, an embodiment of the present application further provides a device for revising clauses, the device comprising:

[0034] The first acquisition module is used to obtain the terms to be reviewed;

[0035] A first matching module is used to determine at least one clause revision record in a clause revision library that matches the clause to be reviewed, wherein the clause revision library includes multiple clause revision records, each of which includes an original clause and revised content for the original clause;

[0036] A revision module is used to revise the clause to be reviewed by using the at least one clause revision record.

[0037] In a possible implementation manner, the first matching module is specifically configured to determine at least one clause revision record in the clause revision library whose similarity with the clause to be reviewed meets a preset condition.

[0038] In a possible implementation manner, the first matching module includes:

[0039] A calculation unit, used to respectively calculate the similarity between the clause to be reviewed and the original clauses included in each clause revision record belonging to the target semantic category in the clause revision library, wherein the semantics of the clause to be reviewed belongs to the target semantic category;

[0040] A first determination unit is used to determine at least one clause revision record from multiple clause revision records belonging to the target semantic category based on the similarity between the clause to be reviewed and the original clauses included in each clause revision record, wherein the similarity between the original clauses included in each clause revision record in the at least one clause revision record and the clause to be reviewed is higher than the similarity between the original clauses included in other clause revision records in the clause revision library and the clause to be reviewed.

[0041] In a possible implementation manner, the clause revision library includes clause revision records of multiple semantic categories, the multiple semantic categories include the target category, and the device further includes:

[0042] A second acquisition module is used to acquire historical revision data, wherein the historical revision data includes clause revision records revised for multiple original clauses within a historical time period;

[0043] A vectorization module, used to vectorize the historical revision data to obtain a vectorized representation corresponding to the historical revision data;

[0044] The first clustering module is used to cluster multiple clause revision records included in the historical revision data according to the vectorized representation corresponding to the historical revision data, and determine the semantic category to which each clause revision record in the historical revision data belongs.

[0045] In a possible implementation, the revision module includes:

[0046] A presenting unit, configured to present a recommendation interface, wherein the recommendation interface includes the at least one clause revision record;

[0047] A second determining unit, configured to determine a target clause revision record from the at least one clause revision record in response to a user's selection operation on the recommendation interface for the clause revision record;

[0048] The first revision unit is used to revise the to-be-reviewed clause according to the revision content in the revision record of the target clause.

[0049] In a possible implementation, the revision module includes:

[0050] A third determining unit is used to determine a target clause revision record from the at least one clause revision record, wherein the original clause in the target clause revision record has the greatest similarity with the clause to be reviewed;

[0051] The second revision unit is used to revise the clause to be reviewed according to the revision content in the revision record of the target clause.

[0052] In a possible implementation, the device further includes:

[0053] A second matching module, used for matching the to-be-reviewed clause with the trust clause in the trust clause library before revising the to-be-reviewed clause;

[0054] Then, the first matching module is used to determine at least one clause revision record matching the clause to be reviewed in the clause revision library when there is no trust clause matching the clause to be reviewed in the trust clause library.

[0055] In a possible implementation, the device further includes:

[0056] A third acquisition module is used to acquire a collection clause library, wherein the collection clause library includes a plurality of candidate clauses;

[0057] A second clustering module is used to cluster the multiple candidate clauses according to the semantics of the multiple candidate clauses to obtain multiple clusters, wherein the semantics of the candidate clauses in each cluster are similar, and the semantics of the candidate clauses in different clusters are different;

[0058] The determination module is used to determine the candidate clauses satisfying the trust condition in each cluster as trust clauses, and add the trust clauses to the trust clause library.

[0059] In a third aspect, an embodiment of the present application further provides a computing device, which may include a processor and a memory:

[0060] The memory is used to store computer programs;

[0061] The processor is used to execute the method described in the first aspect and any one of the embodiments of the first aspect according to the computer program.

[0062] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, characterized in that the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method described in the above-mentioned first aspect and any one of the implementation methods of the first aspect.

[0063] In the above implementation method of the embodiment of the present application, the clause to be reviewed is obtained, and at least one clause revision record matching the clause to be reviewed in the clause revision library is determined, wherein the clause revision library includes multiple clause revision records, and each clause revision record includes the original clause and the revised content of the original clause; then, the clause to be reviewed is revised using the at least one clause revision record determined. In this way, for each clause that the user needs to review, the clause can be automatically reviewed and revised, which can not only effectively reduce labor costs, but also avoid the review results of the clause from being affected by human factors, so as to protect the interests of users as much as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0065] Figure 1 This is a schematic diagram of an exemplary system architecture for revising clauses in an embodiment of the present application;

[0066] Figure 2 This is a flow chart of a method for revising clauses in an embodiment of the present application;

[0067] Figure 3 is a schematic diagram of an exemplary recommendation interface;

[0068] Figure 4 This is a flowchart of another method for revising clauses in an embodiment of the present application;

[0069] Figure 5 This is a schematic diagram of a device structure for revising clauses in an embodiment of the present application;

[0070] Figure 6 A schematic diagram of the hardware structure of a computing device in an embodiment of the present application. DETAILED DESCRIPTION

[0071] See also Figure 1 , is a schematic diagram of the system structure for revising clauses provided in the embodiment of the present application. Figure 1 As shown, the system 100 includes a client 101 and a server 102. There is a communication connection between the client 101 and the server 102, and normal data communication can be performed.

[0072] The client 101 can present an interactive interface to the user, and prompt the user to input the terms to be reviewed (hereinafter referred to as the terms to be reviewed) in the interactive interface. Then, the client 101 can send the terms to be reviewed input by the user to the server 102. After receiving the terms to be reviewed, the server 102 can call a terms revision library storing multiple terms revision records, and determine at least one terms revision record in the terms revision library that matches the terms to be reviewed, wherein each terms revision record includes the original terms and the revised content for the original terms. In this way, the server 102 can revise the terms to be reviewed using the at least one determined terms revision record, and specifically can revise the terms to be reviewed using the revised content included in one of the terms revision records. Finally, the server 102 can return the revised terms to be reviewed to the client 101, so that the client 101 presents the revised terms to be reviewed to the user.

[0073] In this way, for each clause that the user needs to review, the client 101 and the server 102 can be used to automatically review and revise the clause, which can not only effectively reduce labor costs, but also avoid the review results of the clause being affected by human factors, so as to protect the interests of users as much as possible.

[0074] Understandably, Figure 1 The structure of the system 100 for revising clauses shown is only an example provided by the embodiment of the present application. In actual application, the system structure for revising clauses is not limited to Figure 1 Example shown. For example, in other possible implementations, the system for revising terms can provide users with local review terms services. In this case, the system can only include a local user device (such as a local computer, etc.), and the user device executes the terms revision process performed by the above server 102, wherein the user device can also present a corresponding human-computer interaction interface to receive the terms input by the user and present the revised results for the terms to the user. In short, the embodiments of the present application can be applied to any applicable system, and are not limited to the above examples.

[0075] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, various non-limiting implementation methods in the embodiments of the present application are exemplarily described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0076] See also Figure 2 , Figure 2A schematic diagram of a method for revising clauses in an embodiment of the present application is shown. The method can be applied to Figure 1 The system 100 shown in FIG. 1 can, of course, also be applied to other applicable systems. Figure 1 The system 100 shown is described by taking the example of the server 102 performing the review and revision process for the terms to be reviewed. The method may specifically include:

[0077] S201: Obtain the terms to be reviewed.

[0078] In this embodiment, the clauses to be reviewed may be, for example, items in legal texts or agreement documents, such as clauses in a lease contract that stipulate the responsibilities of both parties, clauses in a commercial contract that stipulate the distribution of profits, etc.

[0079] In an exemplary embodiment, a user may provide a file recording one or more clauses, such as a lease contract, a legal document, etc., to the client 101 by means of data copying or file importing. Then, the client 101 may forward the file to the server 102. In this way, after receiving the file, the server 102 may review each clause in the file and revise the problematic clauses accordingly. For ease of understanding and explanation, the following is an example of an exemplary explanation of the process in which the server 102 reviews and revises one of the clauses recorded in the file (i.e., the above-mentioned clause to be reviewed). For the remaining clauses recorded in the file, the revision process for the clause to be reviewed in this embodiment may be referred to for corresponding processing.

[0080] Of course, in other possible implementations, the user may also directly input the terms to be reviewed into the client 101, such as the user may directly input one or more terms on the client 101, and the client 101 sends them to the server 102. In this embodiment, the specific implementation method of how the user provides the terms to be reviewed to the client 101 is not limited.

[0081] S202: Determine at least one clause revision record in a clause revision library that matches the clause to be reviewed, wherein the clause revision library includes multiple clause revision records, and each clause revision record includes an original clause and revised content for the original clause.

[0082] In this embodiment, the server 102 can determine whether the terms to be reviewed need to be revised based on historical data. In specific implementation, the server 102 can create a terms revision library based on multiple terms revision records in advance, so as to use the terms revision records to verify whether the terms to be reviewed need to be revised. Among them, the terms revision record refers to the record of the revision of the terms by the auditor during the manual review of the terms, including the original terms (that is, the terms before manual revision) and the revised content for the original terms. For example, the original terms are specifically "Party A shall pay a deposit of XXX yuan to Party B within 7 days after signing the contract with Party B", and when the auditor revise the original terms, the revised content is specifically to change the "deposit" in the original terms to "deposit". In actual application, after the auditor revises a certain original term, the client 101 or the server 102 can generate a terms revision record including the original terms and the revised content according to the auditor's revision operation on the original term, so as to subsequently establish a terms revision library based on the terms revision record.

[0083] As an implementation example of building a clause revision library, server 102 can first obtain historical revision data, which includes multiple clause revision records generated within a historical time period (such as the past year, etc.), wherein each clause revision record is a record of manual revisions made by the reviewer to an original clause within the historical time period.

[0084] Then, the server 102 can vectorize the historical revision data, and map each clause revision record in the historical revision data to the same vector space, so as to obtain the vectorized representation corresponding to the historical revision data. For example, the server 102 can vectorize each clause revision record in the historical revision data in units of words, such as vectorizing the clause revision record through a bag-of-words model, an n-gram algorithm, a TF-IDF algorithm, a word2vec algorithm, etc.; or, the server 102 can vectorize each clause revision record in the historical revision data in units of sentences, such as vectorizing the clause revision record through a linear discriminant analysis (LDA) algorithm, a long short-term memory (LSTM) algorithm, a BERT (Bidirectional Encoder Representations from Transformers) model, a Sentence-BERT model, etc. Among them, when the Sentence-BERT model is used to vectorize the clause revision record, the semantic features of the sentence can be better captured. Accordingly, the server 102 can extract the hidden layer representation of the [CLS] position in the Sentence-BERT model as the vectorized representation of the clause revision record.

[0085] Finally, the server 102 may cluster the multiple clause revision records included in the historical revision data according to the vectorized representation corresponding to the historical revision data, and determine the semantic category to which each clause revision record in the historical revision data belongs.

[0086] For example, the process of clustering multiple clause revision records according to the vectorized representation of the clause revision records may be:

[0087] (1) The server 102 may treat the vectorized representation corresponding to each clause revision record as a node.

[0088] (2) The server 102 calculates the similarity between different nodes. For example, the server 102 may calculate the vector distance between the vectorized representations corresponding to different nodes, and determine the similarity between the two nodes based on the vector distance between the two nodes. Generally, the smaller the vector distance, the higher the similarity between the two nodes, that is, the more similar the clause revision records corresponding to the two nodes are; conversely, the larger the vector distance, the lower the similarity between the two nodes.

[0089] (3) The server 102 may connect nodes with higher similarity in descending order of similarity, and not connect nodes with lower similarity, such as the server 102 may only connect 10 nodes with the highest similarity each time. The node after the connection is established may be regarded as a new node. At this time, the new node includes multiple clause revision records, and the vectorized representation corresponding to the new node may be obtained by merging the vectorized representations corresponding to the multiple clause revision records. In this way, a tree diagram may be formed based on the multiple nodes, in which some nodes are connected, while other nodes are not connected.

[0090] (4) The server 102 may divide the tree diagram into multiple sub-graphs according to the node connection status in the tree diagram, each sub-graph includes one node or multiple connected nodes, and nodes in different sub-graphs are not connected. In this way, the server 102 completes a round of clustering operation.

[0091] (5) The server 102 may iterate the above clustering operation, i.e., recalculate the similarity between different nodes, and connect some nodes with the highest similarity in descending order (multiple connected nodes can be regarded as a new node), thereby generating a new tree diagram and dividing the new tree diagram to obtain multiple new sub-graphs. It is worth noting that in each round of iterative clustering, the nodes processed by the server 102 are all nodes that have not established connections with other nodes in the previous round of iterative clustering and new nodes generated based on multiple connected nodes.

[0092] The server 102 can perform multiple iterative clustering operations, i.e., iteratively perform (2) to (4), until a preset iterative termination condition is met, which can be, for example, that each subgraph generated by the last iteration includes multiple connected nodes. In this way, the server 102 can eventually obtain multiple subgraphs, each of which corresponds to a semantic category, i.e., the semantics of the clause revision records corresponding to each node in each subgraph all belong to the semantic category. In this way, the server 102 can achieve clustering of multiple clause revision records based on the vectorized representation corresponding to the clause revision records, and can further generate a clause revision library based on the clustered clause revision records.

[0093] After constructing the clause revision library, the server 102 can match the clause revision records in the clause revision library with the clauses to be reviewed. When there is a clause revision record that matches the clause to be reviewed, it indicates that there are auditors who have made revisions to the same or similar clauses to be reviewed in the historical time period. At this time, the server 102 can determine that the clause to be reviewed needs to be revised. When there is no clause revision record that matches the clause to be reviewed, the server 102 can determine that there is currently no revision plan for the clause to be reviewed. In some scenarios, the server 102 can further determine that the clause to be reviewed has passed the review and does not need to be revised.

[0094] It is worth noting that the various clause revision records in the clause revision library have been pre-semantically clustered, and the differences between clause revision records belonging to different semantic categories are usually large (including semantics and clause content). Therefore, in the process of matching the clause revision records in the clause revision library with the clauses to be reviewed, the server 102 may, for example, first identify the semantics of the clauses to be reviewed and determine the target semantic category to which the semantics belongs, so that the server 102 can match the clauses to be reviewed with the clause revision records in the clause revision library belonging to the target semantic category one by one. In this way, the number of clause revision records that are matched with the clauses to be reviewed can be reduced, thereby reducing the computing resource consumption of the server 102.

[0095] As an example, when server 102 determines the clause revision record in the clause revision library that matches the clause to be reviewed, it may be determined based on the similarity between the clause to be reviewed and the original clause. Specifically, it may be determined that at least one clause revision record in the clause revision library whose similarity with the clause to be reviewed meets a preset condition.

[0096] For example, the preset condition may specifically be that the similarity between the determined clause revision record and the clause to be reviewed is greater than the similarity between other clause revision records in the clause revision library and the clause to be reviewed. In specific implementation, the server 102 may respectively calculate the similarity between the clause to be reviewed and the original clause included in each clause revision record belonging to the target semantic category in the clause revision library, such as determining the similarity between the two clauses by calculating the vector distance between the vectorized representation corresponding to the clause to be reviewed and the vectorized representation corresponding to the original clause, or calculating the similarity between the two clauses by a text matching algorithm (the similarity is 100% if the text is completely consistent). Then, the server 102 may determine at least one clause revision record from multiple clause revision records belonging to the target semantic category based on the similarity between the clause to be reviewed and the original clause included in each clause revision record, wherein the similarity between the original clause included in each clause revision record in the determined at least one clause revision record and the clause to be reviewed is higher than the similarity between the original clause included in other clause revision records under the target semantic category and the clause to be reviewed, and is also higher than the similarity between the original clause included in other clause revision records in the clause revision library and the clause to be reviewed. For example, server 102 may determine that the top K (K is a positive integer) clause revision records with the highest similarity match the clause to be reviewed, or server 102 may determine that one or more clause revision records with a similarity greater than a threshold (such as 85%) match the clause to be reviewed. This embodiment does not limit this.

[0097] It is worth noting that the above-mentioned implementation method of constructing a clause revision library and determining the clause revision records that match the clauses to be reviewed is only an exemplary description. In actual application, the server 102 may also construct a clause revision library and determine the clause revision records that match the clauses to be reviewed in other ways. For example, in other possible examples, the server 102 may directly add multiple clause revision records to the clause revision library without clustering the multiple clause revision records; accordingly, when determining the clause revision records that match the clauses to be reviewed, the server 102 may match the clauses to be reviewed with the clause revision records in the clause revision library one by one.

[0098] In actual application, server 102 can regularly update the clause revision library. For example, after a period of time of review and revision of the clauses by the auditors, server 102 can gradually accumulate a certain number of new clause revision records, so that server 102 can merge the new clause revision records and the existing clause revision records in the clause revision library, and can re-perform semantic clustering to update the clause revision library. Of course, server 102 can also create a new clause revision library based on the new clause revision records, etc., and this embodiment does not limit this. In addition, in other embodiments, the execution entity for constructing the clause revision library can also be other devices besides server 102, and after the clause revision library is constructed, the other device sends it to server 102, etc.

[0099] S203: Revise the clause to be reviewed using at least one clause revision record.

[0100] In this embodiment, the server 102 can determine one or more clause revision records from the clause revision library. When there is only one clause revision record that matches the clause to be reviewed, the server 102 can revise the clause to be reviewed according to the clause revision record. When there are multiple clause revision records that match the clause to be reviewed at the same time, the server 102 can filter out one clause revision record from the multiple clause revision records, and revise the clause to be reviewed according to the filtered clause revision record.

[0101] As an implementation example, the server 102 may allow the user to decide which clause revision record to select for revision of the clause to be reviewed. Specifically, the server 102 may present a recommendation interface to the user through the client 101, and the recommendation interface includes one or more clause revision records determined by the server 102, such as Figure 3The clause revision records 1 to 3 are shown. In this way, the user can choose which clause revision record to adopt on the recommendation interface to revise the clause to be reviewed. In actual application, the specific content of the clause to be reviewed can also be presented in the recommendation interface, and a "details" control is also provided in the recommendation interface, so that after the user clicks the "details" control corresponding to one of the clause revision records, the recommendation interface can further present the specific information corresponding to the clause revision record, such as presenting the specific text content of the original clause and the revised content of the original clause, so that the user can determine which clause revision record to select to revise the clause to be reviewed by comparing the text content difference between the clause to be reviewed and the original clause on the recommendation interface, thereby improving the user's experience. Accordingly, the server 102 can respond to the user's selection operation on the recommendation interface for the clause revision record, determine the target clause revision record for revising the clause to be reviewed, and can further revise the clause to be reviewed according to the revised content of the original clause in the target clause revision record.

[0102] For example, assuming that the original clause is "Party A shall pay a deposit of XXX yuan to Party B within 7 days after signing a lease contract with Party B", the revised content is to change the "deposit" in the original clause to "down payment". When the clause to be reviewed is "Party A shall pay a deposit of XXX yuan to Party B within 7 days after signing a commercial contract with Party B", the server 102 can change the "deposit" in the clause to be reviewed to "down payment" according to the revised content, so that the revised clause to be reviewed is: "Party A shall pay a deposit of XXX yuan to Party B within 7 days after signing a commercial contract with Party B".

[0103] Furthermore, for the clause revision record selected by the user, the server 102 can increase the similarity between the original clause in the clause revision record and the clause to be reviewed. In this way, when the server 102 subsequently reviews and revises the same clause to be reviewed input by other users, the clause revision record can be recommended to other users in priority, such as being placed in the primary recommendation position, or the recommended position of the clause revision record is higher than the last recommended position, etc.

[0104] In actual application scenarios, when the auditor manually revises some clauses, the auditor may revise the date, amount, address and other information related to the specific application scenario involved in the clause, while when the auditor manually revises another part of the clause, the revision content is irrelevant to the specific application scenario, such as revising the "deposit" in the clause to "deposit", etc., which is applicable in multiple application scenarios. Therefore, in one example, in the clause revision library constructed by server 102, the clause revision records can be divided into two categories, namely specific revisions and general revisions. Among them, the clause revision records belonging to specific revisions indicate that the revisions made by the auditor to the original clause are related to specific scenarios, such as modifications to the date, amount, and address in the original clause; the clause revision records belonging to general revisions indicate that the revisions made by the auditor to the original clause are irrelevant to specific scenarios, that is, they can be applied to multiple scenarios. In this way, before recommending a clause revision record to a user, the server 102 can first determine whether the revision content included in the clause revision record belongs to a general revision. If so, the recommended clause revision record is presented to the user. If not, the clause revision record is not recommended to the user, such as directly revising the clause to be reviewed according to the clause revision record. Of course, in other possible examples, the server 102 can also filter the clause revision records when constructing the clause revision library, retain only the clause revision records that belong to general revisions, and use this to establish the clause revision library, which is not limited in this embodiment.

[0105] In addition to the above implementation method of selecting the target clause revision record by the user, in other possible implementation examples, the server 102 may also review and revise the clause to be reviewed without user participation. Specifically, the server 102 may automatically determine the target clause revision record from at least one determined clause revision record, such as randomly selecting a clause revision record as the target clause revision record through a random algorithm, or taking the clause revision record with the greatest similarity between the original clause and the clause to be reviewed as the target clause revision record. Then, the server 102 revises the clause to be reviewed based on the revised content of the original clause in the target clause revision record.

[0106] In this way, for each clause that the user needs to review, server 102 can automatically review and revise the clause, which can not only effectively reduce labor costs, but also avoid the review results of the clause being affected by human factors, so as to protect the interests of users as much as possible.

[0107] In the above embodiment, the server 102 uses the clause revision library to revise the clauses to be reviewed as an example for illustration. However, in actual application scenarios, there are usually a large number of clauses in the document that do not need to be revised. Therefore, the server 102 may not need to perform the above revision process on these clauses. Figure 4 , is a flow chart of another method for revising clauses provided in an embodiment of the present application. In this embodiment, the server 102 may first determine whether the clause to be reviewed is a trust clause that does not need to be revised. If not, the server 102 performs a revision process on the clause to be reviewed. Figure 4 As shown, the method may specifically include:

[0108] S401: Obtain the terms to be reviewed.

[0109] S402: Match the terms to be reviewed with the trust terms in the trust terms library.

[0110] In this embodiment, the server 102 may pre-build a trust clause library, which includes multiple trust clauses. A trust clause refers to a high-quality clause with correct content and trustworthiness, and when the clause to be reviewed matches the trust clause, the server 102 may determine that the clause to be reviewed has passed the review without the need for revision.

[0111] Exemplarily, the server 102 can determine the trusted clauses from the clauses reviewed by multiple auditors through adaptive learning. Specifically, the server 102 can first obtain a collection clause library, which includes multiple candidate clauses. The candidate clauses can be, for example, clauses that are correct and reasonable as determined by one or more auditors when manually reviewing the clauses as candidate clauses and added to the collection clause library. Due to the different professional levels of different auditors in actual application and the human factors that have a greater impact on the recognition of clause content, auditors may add some low-quality clauses to the collection clause library when reviewing clauses. Therefore, the server 102 can filter out some candidate clauses from the collection clause library as trusted clauses. For example, the server 102 can cluster the semantics of the multiple candidate clauses according to the semantics of the multiple candidate clauses to obtain multiple clusters, each of which has similar semantics, while the semantics of the candidate clauses in different clusters are different. Among them, the specific implementation method of the server 102 for semantic clustering of multiple candidate clauses can refer to the aforementioned text clustering process, or adopt other usable implementation methods, which are not limited or elaborated in this embodiment. Finally, the server 102 may determine the candidate clauses satisfying the trust condition in each cluster as trust clauses, and add the trust clauses to the trust clause library.

[0112] Exemplarily, when determining whether a candidate clause satisfies the trust condition, the server 102 may specifically present the candidate clause to one or more other auditors for review. When the candidate clause is also determined by other auditors to have passed the review, the server 102 may determine that the candidate clause satisfies the trust condition, otherwise it is determined that it does not meet the trust condition. Alternatively, when determining whether a candidate clause satisfies the trust condition, the server 102 may calculate the average similarity between each candidate clause in the cluster and other candidate clauses for each cluster, and sort each candidate clause according to the average similarity, so that the server 102 may determine the candidate clauses with an average similarity greater than the similarity threshold as trust clauses that meet the trust condition, and determine that the candidate clauses with an average similarity less than the similarity threshold do not meet the trust condition. Alternatively, the server 102 may combine these two implementation methods to filter out trust clauses, such as the server 102 may filter out candidate clauses with an average similarity greater than the similarity threshold in each cluster, and present the filtered candidate conditions to one or more other auditors for review, so that the server 102 may determine the candidate clauses that have passed the review as trust clauses, etc. In this embodiment, the specific implementation method of the server 102 determining whether the candidate clause meets the trust condition is not limited.

[0113] In addition, if in actual application there are multiple reviewers who add the same clause to the collection clause library, the server 102 can also directly determine the clause as a trusted clause.

[0114] In other embodiments, the execution subject of building the trust clause library may also be other devices besides the server 102, and after building the trust clause library, the other device sends it to the server 102, etc.

[0115] After obtaining the trust clause library, the server 102 can determine the trust clause in the trust clause library that matches the clause to be reviewed by exact matching or fuzzy matching. Wherein, when exact matching is adopted, the server 102 can traverse each trust clause in the trust clause library, and if the trust clause is completely consistent with the clause to be reviewed, the server 102 can determine that it matches the clause to be reviewed, and can further determine that the clause to be reviewed has passed the review; otherwise, the server 102 determines that the trust clause does not match the clause to be reviewed. When fuzzy matching is adopted, if the trust clause in the trust clause library is not completely consistent with the clause to be reviewed, but the difference between the trust clause and the clause to be reviewed is only in specific content such as date, amount, address, etc., the server 102 can also determine that it matches the clause to be reviewed, and further determine that the clause to be reviewed has passed the review; and if the difference between the trust clause and the clause to be reviewed includes other aspects of content, the server 102 determines that the trust clause does not match the clause to be reviewed. In actual application scenarios, the server 102 can also present the audit results for the terms to be reviewed, and if the terms to be reviewed pass the review, the server 102 can mark them with a specific color (such as black on a green background, etc.) or other methods, and if the terms to be reviewed fail to pass the review, the server 102 can mark them with other colors or other methods to distinguish between the terms that have passed the review and those that have not passed the review. In actual application, if there are a large number of trust terms in the trust terms library and the expressions of the trust terms are standardized, the server 102 usually has a high hit rate based on the terms of the trust terms library, which can effectively reduce the workload of manual review.

[0116] In this embodiment, when the clause to be reviewed does not match the trust clause in the trust clause library, the server 102 can revise the clause to be reviewed through the clause revision library to improve the content quality of the clause to be reviewed. Alternatively, the server 102 can further review the clause based on the pre-set clause review rules and review elements on the basis of matching with the trust clause library, and when the review is passed, the server 102 can determine that the clause to be reviewed has passed the review and does not need to be revised. When the review is not passed, the server 102 can revise the clause to be reviewed through the clause revision library. At this time, the server 102 can also make risk prompts, such as prompting "missing clauses" (such as incomplete clause content, the entire clause content is missing, etc.), "poor clause quality needs to be replaced" (such as outdated clause content, incoherent sentences, etc.).

[0117] Among them, the audit elements refer to the clause contents that are focused on when auditing the clauses, such as "date", "contact person", "ID number", "address", "payment method", "confidentiality clause", "liability for breach of confidentiality obligation", "amount of liquidated damages", etc. The audit rules refer to the rules that the audit elements must meet, which can usually be formulated by the auditor. For example, for the audit element "ID number", the audit rules defined by the auditor can be: if the ID number has the correct number of digits and the checksum passes the verification, the audit is output as passed; if the audit rules are not met, an alarm will be issued and a prompt will be given.

[0118] S403: When there is no trust clause matching the clause to be reviewed in the trust clause library, determine at least one clause revision record matching the clause to be reviewed in the clause revision library, wherein the clause revision library includes multiple clause revision records, and each clause revision record includes the original clause and the revised content for the original clause.

[0119] S404: Revise the clause to be reviewed using the determined at least one clause revision record.

[0120] In this embodiment, the specific implementation of step S401 and steps S403 and S404 can be found in the relevant description of the above-mentioned embodiment, which will not be described in detail here.

[0121] In this way, by trusting the clause library and the clause revision library, the server 102 can perform a double review and automatic revision on the clauses to be reviewed, thereby improving the accuracy of the clause review and improving the review quality.

[0122] In addition, the present application embodiment also provides a device for revising clauses. Figure 5 , Figure 5 A schematic diagram of a device structure for revising clauses in an embodiment of the present application is shown, and the device 500 includes:

[0123] The first acquisition module 501 is used to acquire the terms to be reviewed;

[0124] A first matching module 502 is used to determine at least one clause revision record in a clause revision library that matches the clause to be reviewed, wherein the clause revision library includes multiple clause revision records, each of which includes an original clause and revised content for the original clause;

[0125] The revision module 503 is used to revise the clause to be reviewed by using the at least one clause revision record.

[0126] In a possible implementation, the first matching module 502 is specifically configured to determine at least one clause revision record in the clause revision library whose similarity with the clause to be reviewed meets a preset condition.

[0127] In a possible implementation manner, the first matching module 502 includes:

[0128] A calculation unit, used to respectively calculate the similarity between the clause to be reviewed and the original clauses included in each clause revision record belonging to the target semantic category in the clause revision library, wherein the semantics of the clause to be reviewed belongs to the target semantic category;

[0129] A first determination unit is used to determine at least one clause revision record from multiple clause revision records belonging to the target semantic category based on the similarity between the clause to be reviewed and the original clauses included in each clause revision record, wherein the similarity between the original clauses included in each clause revision record in the at least one clause revision record and the clause to be reviewed is higher than the similarity between the original clauses included in other clause revision records in the clause revision library and the clause to be reviewed.

[0130] In a possible implementation manner, the clause revision library includes clause revision records of multiple semantic categories, the multiple semantic categories include the target category, and the apparatus 500 further includes:

[0131] A second acquisition module is used to acquire historical revision data, wherein the historical revision data includes clause revision records revised for multiple original clauses within a historical time period;

[0132] A vectorization module, used to vectorize the historical revision data to obtain a vectorized representation corresponding to the historical revision data;

[0133] The first clustering module is used to cluster multiple clause revision records included in the historical revision data according to the vectorized representation corresponding to the historical revision data, and determine the semantic category to which each clause revision record in the historical revision data belongs.

[0134] In a possible implementation, the revision module 503 includes:

[0135] A presenting unit, configured to present a recommendation interface, wherein the recommendation interface includes the at least one clause revision record;

[0136] A second determining unit, configured to determine a target clause revision record from the at least one clause revision record in response to a user's selection operation on the recommendation interface for the clause revision record;

[0137] The first revision unit is used to revise the to-be-reviewed clause according to the revision content in the revision record of the target clause.

[0138] In a possible implementation, the revision module 503 includes:

[0139] A third determining unit is used to determine a target clause revision record from the at least one clause revision record, wherein the original clause in the target clause revision record has the greatest similarity with the clause to be reviewed;

[0140] The second revision unit is used to revise the clause to be reviewed according to the revision content in the revision record of the target clause.

[0141] In a possible implementation manner, the device 500 further includes:

[0142] A second matching module, used for matching the to-be-reviewed clause with the trust clause in the trust clause library before revising the to-be-reviewed clause;

[0143] Then, the first matching module 502 is used to determine at least one clause revision record matching the clause to be reviewed in the clause revision library when there is no trust clause matching the clause to be reviewed in the trust clause library.

[0144] In a possible implementation manner, the device 500 further includes:

[0145] A third acquisition module is used to acquire a collection clause library, wherein the collection clause library includes a plurality of candidate clauses;

[0146] A second clustering module is used to cluster the multiple candidate clauses according to the semantics of the multiple candidate clauses to obtain multiple clusters, wherein the semantics of the candidate clauses in each cluster are similar, and the semantics of the candidate clauses in different clusters are different;

[0147] The determination module is used to determine the candidate clauses satisfying the trust condition in each cluster as trust clauses, and add the trust clauses to the trust clause library.

[0148] It should be noted that the information interaction, execution process, etc. between the modules and units of the above-mentioned device are based on the same concept as the method embodiment in the embodiments of the present application, and the technical effects they bring are the same as those of the method embodiment in the embodiments of the present application. For specific contents, please refer to the description in the method embodiment shown in the aforementioned embodiments of the present application, and will not be repeated here.

[0149] In addition, the present application embodiment also provides a computing device. Figure 6 , Figure 6A schematic diagram of the hardware structure of a computing device in an embodiment of the present application is shown. The device 600 may include a processor 601 and a memory 602.

[0150] Wherein, the memory 602 is used to store computer programs;

[0151] The processor 601 is used to execute the method of revising terms described in the above method embodiment according to the computer program.

[0152] In addition, an embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method of revising terms described in the above method embodiment.

[0153] The "first" in the names of "first acquisition module", "first matching module" and the like mentioned in the embodiments of the present application is only used as a name identifier and does not represent the first in order. The same rule applies to "second", "third" and the like.

[0154] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0155] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, in which the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0156] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A method for amending a clause, characterized in that: The method comprises: Get the terms to be reviewed; Determine at least one clause revision record in a clause revision library that matches the clause to be reviewed, wherein the clause revision library includes multiple clause revision records, and each clause revision record includes an original clause and revised content for the original clause; Determine a target clause revision record from the at least one clause revision record, wherein the original clause in the target clause revision record has the greatest similarity with the clause to be reviewed; The terms to be reviewed shall be revised according to the revision contents in the revision record of the target terms.

2. The method according to claim 1, characterized in that The step of determining at least one clause revision record in the clause revision library that matches the clause to be reviewed includes: Determine at least one clause revision record in the clause revision library whose similarity with the clause to be reviewed meets a preset condition.

3. The method according to claim 2, characterized in that The step of determining at least one clause revision record in the clause revision library whose similarity with the clause to be reviewed meets a preset condition comprises: respectively calculating the similarity between the clause to be reviewed and the original clauses included in each clause revision record belonging to the target semantic category in the clause revision library, wherein the semantics of the clause to be reviewed belongs to the target semantic category; Based on the similarity between the clause to be reviewed and the original clauses included in each clause revision record, at least one clause revision record is determined from multiple clause revision records belonging to the target semantic category, and the similarity between the original clauses included in each clause revision record in the at least one clause revision record and the clause to be reviewed is higher than the similarity between the original clauses included in other clause revision records in the clause revision library and the clause to be reviewed.

4. The method according to claim 3, characterized in that The clause revision library includes clause revision records of multiple semantic categories, wherein the multiple semantic categories include a target category, and the method further includes: Acquire historical revision data, wherein the historical revision data includes revision records of clauses revised for multiple original clauses within a historical time period; Vectorizing the historical revision data to obtain a vectorized representation corresponding to the historical revision data; According to the vectorized representation corresponding to the historical revision data, a plurality of clause revision records included in the historical revision data are clustered to determine the semantic category to which each clause revision record in the historical revision data belongs.

5. The method according to claim 1, characterized in that Before revising the terms to be reviewed, the method further includes: Matching the to-be-reviewed clauses with the trust clauses in the trust clause library; Then, the step of determining at least one clause revision record in the clause revision library that matches the clause to be reviewed includes: When there is no trust clause matching the clause to be reviewed in the trust clause library, at least one clause revision record matching the clause to be reviewed in the clause revision library is determined.

6. The method according to claim 5, characterized in that The method further comprises: Acquire a collection clause library, wherein the collection clause library includes a plurality of candidate clauses; Clustering the multiple candidate clauses according to the semantics of the multiple candidate clauses to obtain multiple clusters, wherein the semantics of the candidate clauses in each cluster are similar, and the semantics of the candidate clauses in different clusters are different; Candidate clauses satisfying the trust condition in each cluster are determined as trust clauses, and the trust clauses are added to the trust clause library.

7. A device for revising clauses, characterized in that: The device comprises: The first acquisition module is used to obtain the terms to be reviewed; A first matching module is used to determine at least one clause revision record in a clause revision library that matches the clause to be reviewed, wherein the clause revision library includes multiple clause revision records, each of which includes an original clause and revised content for the original clause; A revision module is used to determine a target clause revision record from the at least one clause revision record, wherein the original clause in the target clause revision record has the greatest similarity with the clause to be reviewed; and to revise the clause to be reviewed according to the revision content in the target clause revision record.

8. A computing device, characterized in that The device comprises a processor and a memory: The memory is used to store computer programs; The processor is configured to execute the method according to any one of claims 1 to 6 according to the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store a computer program, and the computer program is used to execute the method according to any one of claims 1 to 6.

Images