TWI772975B - Automatic similarity comparison and interpretation method of contracts - Google Patents

Abstract

An automatic similarity comparison and interpretation method of contracts is executed by a computer system that stores a to-be-parsed contract data and multiple reference contract data. The to-be-analyzed contract data includes a plurality of to-be-parsed sentences, each reference contract data includes a plurality of reference sentences, and the method includes the following steps:(A) For each to-be-parsed sentence, the computer system compares the to-be-parsed sentence with the reference sentences to obtain a target reference sentence that is the most similar to the to-be-parsed sentence from the reference sentences in the reference contract data; and (B) For each to-be-parsed sentence, according to the to-be-parsed sentence and the target reference sentence, the computer system uses an parse model for parsing the relevance of the two sentences to generate a parsing result including the relevance of the to-be-parsed sentence and the target reference sentence.

Description

Automatic similarity comparison and analysis method of contract

The present invention relates to an analysis method, in particular to an automatic similarity comparison and analysis method for contracts.

Legal contracts are very important documents in our daily life. A contract is a legal act established by both parties based on the agreement of their wills. It is the main manifestation of the autonomy of private law and stipulates the responsibilities and obligations of both parties. Because of the law involved, creating and signing a contract without careful consideration of its content may risk getting into an unfair or contentious situation and incurring later litigation costs.

However, it is difficult for individual users who lack sufficient legal knowledge to discover the disadvantages of legal issues in contracts and the intent behind contract statements. Even professional lawyers can still spend a lot of time parsing new deeds.

Because a great deal of legal and domain knowledge is required to understand legal statements and spot potential flaws, it is difficult for existing computers to replace human mental activity to parse contracts.

Therefore, the purpose of the present invention is to provide an automatic similarity comparison and analysis method for a contract that can automatically analyze the contract.

Therefore, the automatic similarity comparison and analysis method of the contract of the present invention is performed by a computer system, and the computer system stores a contract data to be analyzed and a plurality of reference contract data, and the to-be-analyzed contract data includes: A plurality of sentences to be parsed, each reference contract document includes a plurality of reference sentences, and the method includes a step (A) and a step (B).

In the step (A), for each sentence to be parsed, the computer system compares the similarity between the sentence to be parsed and the reference sentences of the reference contract data, so as to obtain the data from the reference contract data. A target reference sentence most similar to the to-be-parsed sentence is obtained from the reference sentences.

In the step (B), for each sentence to be parsed, the computer system generates an analytic model including The parsing result of the association between the to-be-parsed statement and the target reference statement.

The effect of the present invention is: for each sentence to be parsed, the computer system compares the similarity between the sentence to be parsed and the reference sentences of the reference contract data to obtain the target reference sentence, and according to The to-be-parsed statement and the target reference statement use the parsing model to generate the parsing result, so that the user can understand the to-be-parsed contract data and the reference based on the parsing result of the pending parsing statement corresponding to the to-be-parsed contract data Correlation between contract documents.

Before the present invention is described in detail, it should be noted that in the following description, similar elements are designated by the same reference numerals.

Referring to FIG. 1, a computer system 1 for implementing the automatic similarity comparison and analysis method of the contract of the present invention is described. The computer system 1 includes a storage unit 11 and a process for electrically connecting the storage unit 11. unit 12. In this embodiment, the implementation of the computer system 1 is, for example, a personal computer, a server or a cloud host, but not limited thereto.

The storage unit 11 stores a plurality of training data, a synonym data including a plurality of synonyms, a contract to be parsed data, and a plurality of reference contract data, each training data includes a first sentence, a second sentence, and a Regarding the indication of the association between the first sentence and the second sentence, the contract document to be parsed includes a plurality of sentences to be parsed, and each reference contract document includes a plurality of reference sentences.

It should be noted that, in this embodiment, the labels of each training data, for example, have six categories, which are the same, the first sentence covers the second sentence, the second sentence covers the first sentence, opposite, irrelevant, and others, These reference contract materials are legally verified and can be divided into 13 categories, namely guidelines regarding user accounts, security clauses regarding privacy and contract information, clauses regarding sufficient conditions for contract formation, defect warranty clauses, dangers Burden of right to rescind the contract and claims of both parties, disclaimers, validity of contracts and third-party suppliers, indemnification or compensation of service providers in the event of returns, returns, definition and description of business, dispute resolution and relief channels, purchasing agents , with respect to jurisdiction and applicable law, and intellectual property rights, without limitation.

Referring to FIGS. 2 and 3 , the embodiment of the automatic similarity comparison and analysis method for contracts of the present invention includes a model building program and a contract analysis program.

Referring to Figures 1 and 2, the model building procedure includes steps 21 to 23, and the steps of the model building procedure are described below.

In step 21, for each training data, the processing unit 12 obtains at least a pair of at least one target word according to at least one target word in the first sentence and the second sentence of the training data and the synonym data synonym group.

In step 22, for each training data, the processing unit 12 generates at least one augmented training data according to the training data and the at least one synonym group.

In step 23 , the processing unit 12 uses a machine learning algorithm to establish a parsing model for parsing the correlation between two sentences according to the training data and the augmented training data generated in step 22 .

For example, the first sentence of the training material is "If you do not agree to the revised terms of a "Service" ________", the second sentence "If _ you do not agree to all or part of these terms ________ , and the training data is marked as the second sentence covering the first sentence. In step 21, if the target word is "agree" and "revise", a synonym group of "agree" including "promise, permission, permission, approval", for example, and "revise" synonym including only "revise", for example, can be obtained group, then in step 22, the processing unit 12 can replace the "agree" of the first sentence and the second sentence with "promise, permission, permission, approval" respectively, to obtain 4 pieces of augmented training data, and also The "revision" of the first sentence can be replaced by "modify" 1 augmented training data, wherein, the label of the augmented training data remains unchanged, and the second sentence covers the first sentence. Similarly, if the target word If it is "Revision", only one augmented training data will be generated.

It is worth noting that in this implementation, the machine learning algorithm is a Natural Language Processing (NLP) pre-training (Neural Network, NN) algorithm, such as BERT (Bidirectional Encoder Representations) from Transformers) model, but not limited to this.

It should be noted again that in this embodiment, since the number of samples for several of the six types of annotations is too low, the data augmentation method of synonym replacement (steps 21 and 22) is used to increase the number of samples. In other embodiments, in the case of sufficient data, data augmentation may not be performed, that is, steps 21 and 22 are not performed, but step 23 is directly performed, and in step 23, the processing unit 12 The analytical model is built based on the training data.

Referring to Figures 1 and 3, the model establishment procedure includes steps 31 and 32, and the steps of the contract parsing procedure are described below.

In step 31, for each to-be-parsed statement, the processing unit 12 compares the similarity between the to-be-parsed statement and the reference statements of the reference contract data, so as to extract the similarity from the reference contract data. A target reference sentence most similar to the to-be-parsed sentence is obtained from the reference sentence.

Referring to FIG. 4 , in the present embodiment, the processing unit 12 uses a local sensitivity hashing (LSH) algorithm to obtain from the reference sentences of the reference contract data a value that is closest to the to-be-parsed sentence Similar to the target reference sentence, step 31 includes sub-steps 311 and 312 , and the sub-steps of step 31 are described below.

In step 311, for each sentence to be parsed, the processing unit 12 uses the locality-sensitive hash algorithm to calculate a plurality of similarities of the similarity of the similarity of the reference sentence related to the sentence to be parsed and the reference contract data, respectively degree value.

It should be noted that, in this embodiment, the processing unit 12 can calculate the to-be-parsed sentence by randomly projecting the to-be-parsed sentence and the reference sentence to be compared into various normalized unit vectors to extract its features. The local-sensitive hash function value of the parsing sentence and these reference sentences. Due to the randomness and size of random projection, the similarity value of the two sentences is the similarity between the features searched according to the hash table, but not limit.

In step 312 , for each sentence to be parsed, the processing unit 12 obtains, according to the similarity values, the target reference sentence with the highest corresponding similarity value from the reference sentences in the reference contract data. .

In step 32, for each sentence to be parsed, the processing unit 12 uses the parsing model to generate an association including the sentence to be parsed and the target reference sentence according to the sentence to be parsed and the target reference sentence corresponding to the sentence to be parsed Sexual analysis results.

Referring to FIG. 5 , step 32 includes sub-steps 321 and 322 , and the sub-steps of step 32 are described below.

In step 321 , for each sentence to be parsed, the processing unit 12 uses a sequence alignment comparison algorithm to align the sentence to be parsed with the target reference sentence to obtain an information related to the sentence to be parsed. Alignment of Parsing Statements The to-be-parsed statement and an alignment reference statement relative to the target reference statement.

In step 322, for each statement to be parsed, the processing unit 12 inputs the aligned statement to be parsed and the aligned reference statement into the parsing model to generate the parsing result.

其中，

代表編輯距離函數(edit distance)， i與 j分別代表該待解析語句與該目標參考語句的對應位置， S1與 S2分別代表該待解析語句與該目標參考語句的字串，

與

分別代表該待解析語句字串 S1在位置 i的字元，及該目標參考語句字串 S2在位置 j的字元。 S1 ＜ S2表示字串 S1較 S2長。 The detailed action of the processing unit 12 using the sequence alignment algorithm is to use the editing (insertion, deletion and replacement) operations of the following formulas to align the to-be-parsed sentence with the candidate reference sentences respectively, so as to compare the The parsed statement and its aligned candidate reference statement are identical, missing, or different parts. d( i , j )=min

in,

represents the edit distance function (edit distance), i and j respectively represent the corresponding positions of the to-be-parsed sentence and the target reference sentence, S 1 and S 2 respectively represent the strings of the to-be-analyzed sentence and the target reference sentence,

and

respectively represent the character at position i of the to-be-parsed sentence string S 1 and the character at position j of the target reference sentence string S 2 . S 1 < S 2 means that the string S 1 is longer than S 2 .

For example, the to-be-parsed statement S1 is, for example, "After you complete the online ordering process, the system will automatically send you a notification via email or other means, but the notification only informs you that the system has received Your order information does not mean that the transaction has been completed or the contract has been established. PChomeOnline reserves the right to accept your order. ”, and the candidate reference sentence S2 compared with the to-be-analyzed sentence is, for example, “After the user completes the online ordering process, it means that an offer is made, and the company will automatically send a notification via email or other means, but the notification is only Indicates that a user order message has been received. ', the processing unit 12 uses the sequence alignment algorithm to obtain results as shown in Table 1 below. Table 1 S1-p1: _After you complete the online ordering procedure, S2-p1: After the user completes the online ordering process, S1-p2: ________ S2-p2: means making an offer, S1-p3: The system will automatically send you a notification via email or other means, S2-p3: The company will automatically send ___ notices by email or other means, S1-p4: But this notification only informs you that the system has received your order information, S2-P4: But this notification is only ____ on behalf of the user's order message has been received, S1-p5: It does not mean that the transaction has been completed or the contract has been established, and PChomeOnline reserves the right to accept your order. S2-p5: _____________________________________.

The statement to be parsed is "_After you complete the online ordering process, _________ will automatically send you a notification by email or other means, but this notification only informs you that the system has received _ you. The order information does not mean that the transaction has been completed or the contract has been established, and PChomeOnline reserves the right to accept your order. ”, the alignment reference sentence is “After the user completes the online ordering process, it means that an offer is made, and the company will automatically send a notice via email or other means to ____, but the notice is only ____ on behalf of having received it. User ordering information, _____________________________________________. '.

It should be noted that, in this embodiment, the first sentence and the second sentence of each training data are also sentences aligned by the processing unit 12 using the sequence alignment algorithm.

To sum up, in the automatic similarity comparison and analysis method of the contract of the present invention, for each sentence to be analyzed, the computer system 1 uses the local sensitive hash to extract the reference sentences from the reference contract data Obtaining the target reference sentence, greatly shortening the comparison time, and inputting the to-be-analyzed sentence and the target reference sentence into the analytical model to generate the analytical result, so that the user can follow the waiting period corresponding to the contract data to be parsed By analyzing the parsing result of the sentence, the relationship between the contract data to be analyzed and the reference contract data can be understood, so the purpose of the present invention can be achieved indeed.

However, the above are only examples of the present invention, and should not limit the scope of implementation of the present invention. Any simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the patent specification are still included in the scope of the present invention. within the scope of the invention patent.

1: Computer system 11: Storage unit 12: Processing unit 21~23: Steps 31, 32: Steps 311, 312: Steps 321, 322: Steps

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, wherein: 1 is a block diagram illustrating a computer system used for an embodiment of the automatic similarity comparison and analysis method of the contract of the present invention; 2 is a flow chart illustrating a model building procedure of this embodiment of the automatic similarity comparison and analysis method of the contract of the present invention; 3 is a flow chart illustrating a contract parsing procedure of this embodiment of the automatic similarity comparison and parsing method for contracts of the present invention; FIG. 4 is a flowchart to assist in explaining the sub-steps of step 31 of the contract parsing procedure of FIG. 3; and FIG. 5 is a flowchart to assist in explaining the sub-steps of step 32 of the contract parsing program of FIG. 3 .

31, 32: Steps

Claims (6)

An automatic similarity comparison and analysis method for contracts is executed by a computer system, the computer system stores a contract data to be analyzed and a plurality of reference contract data, the to-be-analyzed contract data includes a plurality of to-be-analyzed contract data. Parsing sentences, each reference contract document includes a plurality of reference sentences, and the method includes the following steps: (A) For each statement to be parsed, compare the similarity between the statement to be parsed and the reference statements of the reference contract data, so as to obtain a match between the reference statements of the reference contract data The target reference sentence to which the sentence to be parsed is most similar; and (B) For each statement to be parsed, according to the statement to be parsed and the target reference statement corresponding to the statement to be parsed, a parsing model for parsing the correlation between the two statements is used to generate a statement including the statement to be parsed and the target reference statement The result of parsing the associativity of the statement. The method for automatic similarity comparison and analysis of contract documents according to claim 1, wherein, in step (A), for each sentence to be analyzed, according to the sentence to be analyzed and the reference contract data For the reference sentence, a locality-sensitive hash algorithm is used to obtain a target reference sentence most similar to the to-be-parsed sentence from the reference sentences of the reference contract data. The method for automatic similarity comparison and analysis of contract documents as claimed in claim 2, wherein step (A) includes the following sub-steps: (A-1) for each statement to be parsed, using the locality-sensitive hash algorithm to calculate a plurality of similarity values related to the similarity of the statement to be parsed and the reference statement data of the reference contract documents; and (A-2) For each sentence to be parsed, according to the similarity values, obtain the target reference sentence with the highest corresponding similarity value from the reference sentences in the reference contract data. The method for automatic similarity comparison and analysis of contract documents as claimed in claim 1, wherein step (B) includes the following sub-steps: (B-1) For each sentence to be parsed, use a sequence alignment algorithm to align the sentence to be parsed with the target reference sentence to obtain an aligned sentence to be parsed and an alignment related to the sentence to be parsed an aligned reference statement relative to the target reference statement; and (B-2) For each to-be-parsed statement, input the aligned to-be-parsed statement and the aligned reference statement into the parsing model to generate the parsing result. According to the method for automatic similarity comparison and analysis of contract documents as described in claim 1, the computer system further stores a plurality of training data, each training data includes a first sentence composed of a plurality of single characters, a first sentence composed of a plurality of single characters The second sentence composed of and different from the first sentence, and a label related to the relevance of the first sentence and the second sentence, before step (B), it also includes the following steps: (C) Using a machine learning algorithm to build the analytical model according to the training data. The automatic similarity comparison and analysis method of the contract document according to claim 5, the computer system also stores a synonym data including a plurality of synonyms, wherein step (C) includes the following sub-steps: (C-1) For each training data, according to at least one target word in the first sentence and the second sentence of the training data, and the synonym data, obtain at least a pair of synonym groups corresponding to at least one target word; (C-2) For each training data, generate at least one augmented training data according to the training data and the at least one synonym group; and (C-3) According to the training data and the augmented training data generated in step (C-2), use the machine learning algorithm to establish the analytical model.

Images