Text information intelligent analysis method and system based on large language model

Abstract

The invention discloses a text information intelligent analysis method and system based on a large language model, and relates to the technical field of text analysis, legal clauses similar to text information are obtained through an algorithm of text information and legal clause calling probability, and the legal clauses matched through the algorithm are screened manually, so that the legal clauses are obtained. The method comprises the following steps: acquiring legal terms similar to text information through a probability calling algorithm, manually screening the legal terms matched by the algorithm, comparing differences between the manually screened legal terms and the legal terms with the maximum calling probability, extracting difference features of the differences, and extracting the legal terms with the maximum calling probability. And adjusting the recommendation sequence of the legal clauses with the same difference characteristics in the text information.

Description

Technical Field

[0001] The present invention relates to the technical field of text analysis, and in particular to a text information intelligent analysis method and system based on a large language model. Background Art

[0002] In order to reduce the workload in the process of consulting legal documents, a legal knowledge database is established. The method of providing relevant legal knowledge for text content through natural language processing combined with machine learning is gradually promoted. For example, the Chinese patent with publication number: CN111694945A, named "Law Article Association Recommendation Method and Device Based on Neural Network", discloses a method for providing legal document references based on sentence call probability.

[0003] Since the target technology cannot completely get rid of the step of manually verifying the legal recommendation clauses generated by the computer, it is necessary to generalize the manual verification results of the local text to similar positions in the text. The existing technology lacks the means to repeat and generalize the manual verification process. When relevant personnel verify the legal clauses, they still need to compare the results of the calculation recommendations one by one to obtain the most appropriate legal clauses. Summary of the invention

[0004] The purpose of the present invention is to provide a text information intelligent analysis method and system based on a large language model to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a text information intelligent analysis method based on a large language model;

[0006] Step S100: Collecting the annotation records of the text information by relevant personnel, the annotation records include: collecting a certain paragraph in the text information and the legal clauses cited in the certain paragraph, identifying the text elements in any paragraph, combining the text elements in the same paragraph and the corresponding legal clauses into a sample record, and collecting all the sample records of the first target record to obtain a comparison database;

[0007] Step S200: Identify at least two text elements in the text information, combine one of the text elements with the corresponding legal clause into a tuple, calculate the occurrence probability of the other text element under the condition that a certain tuple appears in the comparison database, and record the occurrence probability as the influence coefficient of the other text element on the legal clause;

[0008] Step S300: collecting a certain text information as a target text, inputting each paragraph in the target text into the legal clause calling model, and obtaining an initial recommended sequence of legal clauses corresponding to each paragraph;

[0009] Step S400: taking a paragraph corresponding to a second target record as a first target paragraph, and obtaining the calling probability of two legal clauses in the second target record;

[0010] Step S500: Identify the text elements in the first target paragraph, respectively form two tuples of the same text elements in the legal clause selected by the relevant personnel and the legal clause with the highest call probability, respectively calculate the influence coefficients of the same text elements on the two legal clauses, and extract the difference features between the text elements;

[0011] Step S600: A paragraph that is semantically associated with the first target paragraph is used as a second target paragraph. In the second target paragraph, the order of legal clauses recommended by the second target paragraph is adjusted according to the features of the elements in the legal clauses selected by the relevant personnel.

[0012] Further, step S200 includes:

[0013] Step S201: Identify two text elements in the same paragraph in the text information, record them as the first text element and the second text element respectively, and form a tuple of any text element and the legal clause corresponding to the paragraph, wherein the tuple composed of the second text element and the corresponding legal clause is recorded as (A, C), wherein A represents the second text element, and C represents the legal clause corresponding to the same paragraph;

[0014] Step S202: in the comparison database, the occurrence probability of the bigram (A, C) is calculated, and the occurrence probability pE of the first text element under the condition that the bigram (A, C) appears is calculated, pE=P(E|(A, C)), where E represents the first text element, and P() represents the probability calculation formula;

[0015] Step S203: Obtain a tuple (E, C) consisting of the first text element and the legal clause corresponding to the first target segment, and calculate the probability pA of the second text element appearing under the condition that the tuple (E, C) appears, pA=P(A|(E, C));

[0016] Step S204: record pE as the influence coefficient of the first text element on the legal clause corresponding to the same paragraph, and record pA as the influence coefficient of the second text element on the legal clause corresponding to the same paragraph;

[0017] Taking the occurrence of the binary pair (A, C) as an event, P(E|(A, C)) calculates the probability of the occurrence of event E under the condition that event (A, C) occurs. Taking the occurrence of the binary pair (E, C) as an event, P(A|(E, C)) calculates the probability of the occurrence of event A under the condition that event (E, C) occurs.

[0018] Compare the values ​​of pA and pE to obtain the correlation between the first text element and the second text element and C. When pA>pE, it means that in the historical records, the first text element has a higher correlation with the legal article C than the second text element. The correlation degree is used as a measure of the distance between the text element and the legal article in this scheme.

[0019] Due to the different dominant text elements, the applicable legal provisions in the text information are also different. By calculating the dominant position of different text elements in the text relative to the legal provisions, the recommended legal provisions are optimized. For the same legal provision cla, when P(E|(A, C))>P(A|(E, C)), it means that event E is dominant for the legal provision cla.

[0020] Furthermore, step S300 includes:

[0021] Step S301: obtaining a number of legal clauses matched by the first target paragraph, and obtaining the call probability of each legal clause;

[0022] Step S302: Arrange the call probabilities of the legal clauses and the first target paragraph from high to low, and obtain an initial recommendation sequence of the legal clauses corresponding to the first target paragraph;

[0023] Step S303: obtaining the legal clause with the highest call probability in the initial recommendation sequence and recording it as the first target clause of the first target paragraph, and obtaining the call probability of the legal clause in the first target paragraph and recording it as α.

[0024] Furthermore, step S400 includes:

[0025] Step S401: In the second adjustment record, the legal clause selected by the relevant personnel is recorded as the second target clause of the first target paragraph;

[0026] Step S402: In the first target paragraph, obtain the call probability of the second target clause and record it as β, satisfying the condition β<α.

[0027] Furthermore, step S500 includes:

[0028] Step S501: Obtain the first text element and the second text element in the first target paragraph. Under the condition of the first target clause, the influence coefficient of the first text element is recorded as pE1, and the influence coefficient of the second text element is recorded as pA1. Under the condition of the second target clause, the influence coefficient of the first text element is recorded as pE2. The influence coefficient of the second text element is recorded as pA2.

[0029] Step S502: when pE1>pA1 and pA2>pE2, it is the first adjustment mode. In the first adjustment mode, kd=pA2-pE2 is calculated;

[0030] In the first target paragraph, there is a relationship pE1>pA1. For the first target paragraph, the legal provisions matched by the computer are dominated by the first text element. However, after proofreading by relevant personnel, the adopted legal provisions have a relationship pA2>pE2, which is dominated by the second text element.

[0031] It indicates that in the process of recommending legal clauses for the first target paragraph, there are defects in recommending legal clauses dominated by the first text element and the second text element. In a paragraph with both the first text element and the second text element, the weight of the legal clause dominated by the second text element needs to be increased so that the legal clause dominated by the second text element can be more easily viewed by relevant personnel.

[0032] Step S503: when pE1>pA1, calculate k1=pE1-pA1, k2=pE2-pA2, wherein when k1<k2, it matches the second adjustment mode, and when k1>k2, it matches the third adjustment mode;

[0033] For the first target paragraph, k1 represents the difference between the legal clause matched by the computer and the first text element and the second text element in the first target paragraph, and k2 represents the difference between the legal clause selected by the relevant personnel and the first text element and the second text element in the first target paragraph;

[0034] k1<k2 indicates that the difference in dominance between the first text element and the second text element increases. Compared with the first target paragraph, the difference in dominance between the first text element and the second text element increases in the legal clauses before and after the target adjustment. It can be inferred that the legal clauses with a larger difference in dominance between the first text element and the second text element are more suitable for the target text. It is necessary to increase the weight value of the legal clauses with a larger difference in dominance between the first text element and the second text element in the legal clauses matched to the target text;

[0035] k1>k2 indicates that the gap in dominance between the first text element and the second text element is reduced. Compared with the first target paragraph, the gap in dominance between the first text element and the second text element is reduced in the legal clauses before and after the target adjustment. It can be inferred that the legal clauses with a smaller gap in dominance between the first text element and the second text element are more suitable for the target text. It is necessary to increase the weight values ​​of the legal clauses with a smaller gap in dominance between the first text element and the second text element among the legal clauses matched to the target text.

[0036] Furthermore, step S600 includes:

[0037] Step S601: Obtain an initial recommendation sequence of legal clauses corresponding to the second target paragraph, and establish an adjustment coefficient calculation function H of the legal clauses, H=μ+W, where μ represents the call probability of the legal clauses and the second target paragraph, and W represents the adjustment weight;

[0038] Step S602: Obtain the influence coefficient of the first text element in the second target paragraph, record it as pE3, and the influence coefficient of the second text element, record it as pA3;

[0039] Step S603: acquiring corresponding legal clauses in the initial recommendation sequence of the second target paragraph according to the matching conditions of the first adjustment mode, the second adjustment mode, and the third adjustment mode respectively;

[0040] Step S604: When pE1>pA1, pA2>pE2 and pA3>pE3, match the first adjustment mode, obtain the legal terms that meet the condition pA3>pE3 in the legal terms corresponding to the second target paragraph, calculate ke=pA3-pE3, obtain the call probability μ1 of the legal terms and the second target paragraph, and calculate the adjustment coefficient H1 under the first adjustment mode, H1=μ1+(ke / kd)×(α-β);

[0041] When pE1>pA1, pA3<pE3, and k1<k2, match the second adjustment mode, obtain the legal terms that meet the condition of pA3<pE3 in the legal terms corresponding to the second target paragraph, obtain the call probability μ2 of the legal terms and the second target paragraph, and calculate the adjustment coefficient H2 under the second matching mode, H2=μ2+(k3 / k2)×(α-β), where k3=pE3-pA3;

[0042] When pE1>pA1, pA3<pE3, and k1>k2, match the third adjustment mode, obtain the legal terms that meet the condition of pA3<pE3 in the legal terms corresponding to the second target paragraph, obtain the call probability μ3 of the legal terms and the third target paragraph, and calculate the adjustment coefficient H3 under the third adjustment mode, H3=μ3+(k2 / k3)×(α-β), where k3=pE3-pA3;

[0043] Step S605: In the initial recommendation sequence of the legal terms corresponding to the second target paragraph, the adjustment coefficient is used to replace the call probability of the corresponding legal terms, and the values ​​in the initial recommendation sequence are rearranged from high to low to obtain a recommended update sequence. The legal terms in the recommended update sequence are arranged from high to low according to the call probability or adjustment coefficient corresponding to the legal terms, and the legal terms are recommended to relevant personnel.

[0044] In order to better implement the above method, a text information intelligent analysis system based on a large language model is also proposed, which includes a comparison database management module, an influence coefficient calculation module, an association matching module, a target adjustment management module, a difference calculation module and an adjustment management module, wherein the comparison database management module is used to manage the comparison database, the influence coefficient calculation module is used to calculate the influence coefficient of the text element, the association matching module is used to obtain the legal clauses matching each paragraph, the target adjustment management module is used to record the target adjustment, the difference calculation module is used to calculate the difference degree of the text elements corresponding to the legal clauses before and after the target adjustment, and the adjustment management module is used to adjust the order of the legal clauses matching the second target paragraph;

[0045] Furthermore, the influence coefficient calculation module includes: a text element acquisition unit, a bigram management unit and an influence coefficient calculation unit, wherein the text element acquisition unit is used to acquire text elements in the text information, the bigram management unit is used to manage bigrams consisting of the first text element or the second text element and the corresponding legal clause, and the first influence coefficient calculation unit is used to calculate the influence coefficient of the second text element of the first text element in the first target paragraph;

[0046] Furthermore, the association matching module includes: an association matching model management unit and a first sorting unit, wherein the association matching model management unit is used to manage the legal terms calling model, and the first sorting unit is used to obtain an initial recommended sequence of paragraphs in the target text;

[0047] Further, the target adjustment management module includes: a target adjustment acquisition unit and a call probability comparison unit, wherein the target adjustment acquisition unit is used to acquire the target adjustment, and the call probability comparison unit is used to compare the first target clause and the second target clause of the first target paragraph;

[0048] Furthermore, the difference calculation module includes: an impact difference calculation unit and an adjustment mode acquisition unit, wherein the impact difference calculation unit is used to calculate the difference of impact coefficients under different legal terms in the first target paragraph, and the adjustment mode acquisition unit is used to acquire the adjustment mode in the second target paragraph according to the difference of impact coefficients;

[0049] Furthermore, the adjustment management module includes: a second influence coefficient calculation unit, a pattern matching unit, an adjustment coefficient calculation unit and a second sorting unit, wherein the second influence coefficient calculation unit is used to calculate the influence coefficient of the second text element of the first text element in the second target paragraph, the pattern matching unit is used to match the adjustment pattern corresponding to each legal clause in the legal clause corresponding to the second target paragraph according to the adjustment pattern, the adjustment coefficient calculation unit is used to calculate the adjustment coefficient under each matching pattern using the adjustment coefficient calculation function, and the second sorting unit is used to update the recommended sequence of the legal provisions.

[0050] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention compares the legal clauses selected by humans with the legal clauses recommended by computers, analyzes the dominant text elements in the legal clauses, and analyzes the relative differences of the dominant text elements in the legal clauses after human selection. The changes in the relative differences are used as modification features, and the modification features are repeated and propagated, and the order of the legal clauses recommended by computers is rearranged, so that relevant personnel can more easily find legal clauses that are suitable for the text. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0052] Figure 1 It is a structural schematic diagram of a text information intelligent analysis system based on a large language model of the present invention;

[0053] Figure 2 It is a flow chart of a text information intelligent analysis method based on a large language model of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 and Figure 2 ,The present invention provides a technical solution: a text information intelligent analysis method based on a large language model;

[0056] Step S100: Collecting the annotation records of the text information by relevant personnel, the annotation records include: collecting a certain paragraph in the text information and the legal clauses cited in the certain paragraph, identifying the text elements in any paragraph, combining the text elements in the same paragraph and the corresponding legal clauses into a sample record, and collecting all the sample records of the first target record to obtain a comparison database;

[0057] Specifically, by collecting relevant people and annotations of a certain paragraph, the annotations include (G, cla), where G includes text elements appearing in a certain paragraph, and cla represents the legal terms corresponding to a certain paragraph.

[0058] Step S200: Identify at least two text elements in the text information, combine one of the text elements with the corresponding legal clause into a tuple, calculate the occurrence probability of the other text element under the condition that a certain tuple appears in the comparison database, and record the occurrence probability as the influence coefficient of the other text element on the legal clause;

[0059] In the embodiment, the text elements include: the parties to the contract, the domiciles of the parties, the subject matter of the contract, the place of performance of the contract or the time of performance of the contract; the method of performance of the contract, the obligations specified in the contract or the limitations on breach of contract; the remuneration obtained, the liability for breach of contract or the method of resolving disputes;

[0060] Wherein, step S200 includes:

[0061] Step S201: Identify two text elements in the same paragraph in the text information, record them as the first text element and the second text element respectively, and form a tuple of any text element and the legal clause corresponding to the paragraph, wherein the tuple composed of the second text element and the corresponding legal clause is recorded as (A, C), wherein A represents the second text element, and C represents the legal clause corresponding to the same paragraph;

[0062] Step S202: in the comparison database, the occurrence probability of the bigram (A, C) is calculated, and the occurrence probability pE of the first text element under the condition that the bigram (A, C) appears is calculated, pE=P(E|(A, C)), where E represents the first text element, and P() represents the probability calculation formula;

[0063] Step S203: Obtain a tuple (E, C) consisting of the first text element and the legal clause corresponding to the first target segment, and calculate the probability pA of the second text element appearing under the condition that the tuple (E, C) appears, pA=P(A|(E, C));

[0064] Step S204: record pE as the influence coefficient of the first text element on the legal clause corresponding to the same paragraph, and record pA as the influence coefficient of the second text element on the legal clause corresponding to the same paragraph.

[0065] Step S300: collecting a certain text information as a target text, inputting each paragraph in the target text into the legal clause calling model, and obtaining an initial recommended sequence of legal clauses corresponding to each paragraph;

[0066] Wherein, step S300 includes:

[0067] Step S301: obtaining a number of legal clauses matched by the first target paragraph, and obtaining the call probability of each legal clause;

[0068] Step S302: Arrange the call probabilities of the legal clauses and the first target paragraph from high to low, and obtain an initial recommendation sequence of the legal clauses corresponding to the first target paragraph;

[0069] Step S303: obtaining the legal clause with the highest call probability in the initial recommendation sequence and recording it as the first target clause of the first target paragraph, and obtaining the call probability of the legal clause in the first target paragraph and recording it as α.

[0070] Step S400: taking a paragraph corresponding to a second target record as a first target paragraph, and obtaining the calling probability of two legal clauses in the second target record;

[0071] Wherein, step S400 includes:

[0072] Step S401: In the second adjustment record, the legal clause selected by the relevant personnel is recorded as the second target clause of the first target paragraph;

[0073] Step S402: In the first target paragraph, obtain the call probability of the second target clause and record it as β, satisfying the condition β<α.

[0074] Step S500: Identify the text elements in the first target paragraph, respectively form two tuples of the same text elements in the legal clause selected by the relevant personnel and the legal clause with the highest call probability, respectively calculate the influence coefficients of the same text elements on the two legal clauses, and extract the difference features between the text elements;

[0075] Wherein, step S500 includes:

[0076] Step S501: Obtain the first text element and the second text element in the first target paragraph. Under the condition of the first target clause, the influence coefficient of the first text element is recorded as pE1, and the influence coefficient of the second text element is recorded as pA1. Under the condition of the second target clause, the influence coefficient of the first text element is recorded as pE2. The influence coefficient of the second text element is recorded as pA2.

[0077] Step S502: when pE1>pA1 and pA2>pE2, it is the first adjustment mode. In the first adjustment mode, kd=pA2-pE2 is calculated;

[0078] Step S503: When pE1>pA1, calculate k1=pE1-pA1, k2=pE2-pA2, wherein when k1<k2, it matches the second adjustment mode, and when k1>k2, it matches the third adjustment mode.

[0079] Step S600: obtaining a paragraph that is semantically related to the first target paragraph as a second target paragraph, and adjusting the order of legal clauses recommended for the second target paragraph according to the features of the elements in the legal clauses selected by the relevant personnel in the second target paragraph;

[0080] In the implementation process, the second target paragraph related to the first target paragraph is extracted by extracting related words, text indexing or semantic call probability calculation;

[0081] Wherein, step S600 includes:

[0082] Step S601: Obtain an initial recommendation sequence of legal clauses corresponding to the second target paragraph, and establish an adjustment coefficient calculation function H of the legal clauses, H=μ+W, where μ represents the call probability of the legal clauses and the second target paragraph, and W represents the adjustment weight;

[0083] Step S602: Obtain the influence coefficient of the first text element in the second target paragraph, record it as pE3, and the influence coefficient of the second text element, record it as pA3;

[0084] Step S603: acquiring corresponding legal clauses in the initial recommendation sequence of the second target paragraph according to the matching conditions of the first adjustment mode, the second adjustment mode, and the third adjustment mode respectively;

[0085] Step S604: When pE1>pA1, pA2>pE2 and pA3>pE3, match the first adjustment mode, obtain the legal terms that meet the condition pA3>pE3 in the legal terms corresponding to the second target paragraph, calculate ke=pA3-pE3, obtain the call probability μ1 of the legal terms and the second target paragraph, and calculate the adjustment coefficient H1 under the first adjustment mode, H1=μ1+(ke / kd)×(α-β);

[0086] When pE1>pA1, pA3<pE3, and k1<k2, match the second adjustment mode, obtain the legal terms that meet the condition of pA3<pE3 in the legal terms corresponding to the second target paragraph, obtain the call probability μ2 of the legal terms and the second target paragraph, and calculate the adjustment coefficient H2 under the second matching mode, H2=μ2+(k3 / k2)×(α-β), where k3=pE3-pA3;

[0087] When pE1>pA1, pA3<pE3, and k1>k2, match the third adjustment mode, obtain the legal terms that meet the condition of pA3<pE3 in the legal terms corresponding to the second target paragraph, obtain the call probability μ3 of the legal terms and the third target paragraph, and calculate the adjustment coefficient H3 under the third adjustment mode, H3=μ3+(k2 / k3)×(α-β), where k3=pE3-pA3;

[0088] For example: obtain the first target paragraph L1, obtain the legal clauses cla1 and cla2 that match L1, where the calling probability of cla1 and the first target paragraph is α=0.75, and the calling probability of cla2 and the first target paragraph is β=0.5;

[0089] Obtaining a second target paragraph L2 and legal terms matched by L2;

[0090] In the first adjustment mode, the influence coefficients of the first text element and the second text element in the first target paragraph are calculated, and the conditions pE1>pA1 and pA2>pE2 are met. In the second target paragraph, the legal clauses corresponding to the conditions pA3>pE3 are searched;

[0091] Obtain the legal clause cla31 that meets the conditions among the legal clauses matched by L2, and the calling probability of cla31 and L2 is μ1;

[0092] Calculate the adjustment coefficient H11 = μ1 + (ke / kd) × (α-β). If the boundary condition is met, the difference between the dominant text elements in cla31 and the dominant text elements in cla2 is the same, and the call probability with the second target paragraph is equal to β, that is, μ1 = β, pA3 = pA2, pE3 = pE2, then H11 = α is calculated. If ke>kd, that is, pA3>pE3, the dominant position of the second text element in cla3 is more significant. Under the condition that the call probability remains unchanged, H12 is calculated, H12>H11.

[0093] In the second adjustment mode, the influence coefficients of the first text element and the second text element in the first target paragraph are calculated, and the conditions pE1>pA1 and k1<k2 are met. In the second target paragraph, the legal clauses corresponding to the conditions pE3>pA3 are searched;

[0094] Obtain the legal clause cla32 that meets the conditions among the legal clauses matched by L2, and the calling probability of cla32 and L2 is μ2;

[0095] Calculate the adjustment coefficient H21 = μ2 + (k3 / k2) × (α-β). If the boundary condition is met, the difference between the dominant text elements in cla32 is the same as the difference between the dominant text elements in cla2, and the call probability with the second target paragraph is equal to β, that is, μ2 = β, pA3 = pA2, pE3 = pE2, then H21 = α is calculated. If k3> k2, that is, pE3-pA3> pE3-pA3, in cla3, the difference in the dominant position between the first text element and the second text element is more significant. Under the condition that the call probability remains unchanged, H22 is calculated, H22> H21.

[0096] In the third adjustment mode, the influence coefficients of the first text element and the second text element in the first target paragraph are calculated, and the conditions pE1>pA1 and k1>k2 are met. In the second target paragraph, the legal clauses corresponding to the conditions pE3<pA3 are searched;

[0097] Obtain the legal clause cla33 that meets the conditions among the legal clauses matched by L2, and the calling probability of cla33 and L2 is μ3;

[0098] Calculate the adjustment coefficient H31 = μ3 + (k2 / k3) × (α-β). If the boundary condition is met, the difference between the dominant text elements in cla32 and the dominant text elements in cla2 is the same, and the call probability with the second target paragraph is equal to β, that is, μ3 = β, pA3 = pA2, pE3 = pE2, then H31 = α is calculated. If k3 < k2, that is, pE3-pA3 < pE3-pA3, in cla3, the difference in the dominant position between the first text element and the second text element is reduced, and H32 is calculated under the condition that the call probability remains unchanged, H32 > H31.

[0099] Step S605: In the initial recommendation sequence of the legal terms corresponding to the second target paragraph, the adjustment coefficient is used to replace the call probability of the corresponding legal terms, and the values ​​in the initial recommendation sequence are rearranged from high to low to obtain a recommended update sequence. The legal terms in the recommended update sequence are arranged from high to low according to the call probability or adjustment coefficient corresponding to the legal terms, and the legal terms are recommended to relevant personnel.

[0100] The system includes: a comparison database management module, an influence coefficient calculation module, an association matching module, a target adjustment management module, a difference calculation module and an adjustment management module;

[0101] Wherein, the comparison database management module is used to manage the comparison database;

[0102] The influence coefficient calculation module is used to calculate the influence coefficient of the text element, wherein the influence coefficient calculation module includes: a text element acquisition unit, a bigram management unit and an influence coefficient calculation unit, wherein the text element acquisition unit is used to acquire the text element in the text information, the bigram management unit is used to manage the bigram composed of the first text element or the second text element and the corresponding legal clause, and the first influence coefficient calculation unit is used to calculate the influence coefficient of the second text element of the first text element in the first target paragraph;

[0103] The association matching module includes: an association matching model management unit and a first sorting unit, wherein the association matching model management unit is used to manage the legal clause call model, and the first sorting unit is used to obtain an initial recommended sequence of paragraphs in the target text;

[0104] The target adjustment management module includes: a target adjustment acquisition unit and a call probability comparison unit, wherein the target adjustment acquisition unit is used to acquire the target adjustment, and the call probability comparison unit is used to compare the first target clause and the second target clause of the first target paragraph;

[0105] The difference calculation module includes: an impact difference calculation unit and an adjustment mode acquisition unit, wherein the impact difference calculation unit is used to calculate the difference of impact coefficients under different legal terms in the first target paragraph, and the adjustment mode acquisition unit is used to acquire the adjustment mode in the second target paragraph according to the difference of impact coefficients;

[0106] Among them, the adjustment management module is used to adjust the order of legal clauses matching the second target paragraph, wherein the adjustment management module includes: a second influence coefficient calculation unit, a pattern matching unit, an adjustment coefficient calculation unit and a second sorting unit, wherein the second influence coefficient calculation unit is used to calculate the influence coefficient of the second text element of the first text element in the second target paragraph, the pattern matching unit is used to match the adjustment pattern corresponding to each legal clause in the legal clause corresponding to the second target paragraph according to the adjustment pattern, the adjustment coefficient calculation unit is used to calculate the adjustment coefficient under each matching pattern using the adjustment coefficient calculation function, and the second sorting unit is used to update the recommended sequence of legal provisions.

[0107] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article or device.

[0108] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A text information intelligent analysis method based on a large language model, characterized in that: The method comprises: Organize the legal clause call records of all historical text information, build a legal clause call model for text information through deep learning algorithms, monitor the legal clauses output after inputting text information into the legal clause call model, and the call probability corresponding to the legal clauses, and arrange the output legal clauses and the call probability of the text information from high to low to obtain the initial recommended sequence of legal clauses; Collect the annotation records of the text information by the relevant personnel. When the relevant personnel select the legal clause with the highest probability of being called as the applicable legal clause of the text information, it is recorded as the first target record, and several first target records are collected to form a comparison database; when the legal clause selected by the relevant personnel is not the legal clause with the highest probability of being called as the applicable legal clause of the text information, it is recorded as the second target record; Extract the text elements of the text information in the second target record, form two tuples with the legal clauses selected by the relevant personnel and the legal clauses with the highest probability of being called, and calculate the occurrence probability of the two tuples in the comparison database as the influence coefficient of the text elements on the legal clauses; In the second target record, the influence coefficients of the text elements and the legal clauses selected by relevant personnel and the legal clauses with the highest probability of being called are calculated respectively, and the difference between the two influence coefficients is used as the target feature. Feature extraction is performed on the text elements in the text information, and text elements with the same features as the target features are screened out. The comparative weights of the legal clauses corresponding to the text elements are calculated, and the order of the legal clauses in the initial recommendation sequence is adjusted.

2. The text information intelligent analysis method based on a large language model according to claim 1, characterized in that: The method comprises the following steps: Step S100: Collecting the annotation records of the text information by relevant personnel, the annotation records include: collecting a certain paragraph in the text information and the legal clauses cited in the certain paragraph, identifying the text elements in any paragraph, combining the text elements in the same paragraph and the corresponding legal clauses into a sample record, and collecting all the sample records of the first target record to obtain a comparison database; Step S200: Identify at least two text elements in the text information, combine one of the text elements with the corresponding legal clause into a tuple, calculate the occurrence probability of the other text element under the condition that a certain tuple appears in the comparison database, and record the occurrence probability as the influence coefficient of the other text element on the legal clause; Step S300: collecting a certain text information as a target text, inputting each paragraph in the target text into the legal clause calling model, and obtaining an initial recommended sequence of legal clauses corresponding to each paragraph; Step S400: taking a paragraph corresponding to a second target record as a first target paragraph, and obtaining the calling probability of two legal clauses in the second target record; Step S500: Identify the text elements in the first target paragraph, respectively form two tuples of the same text elements in the legal clause selected by the relevant personnel and the legal clause with the highest call probability, respectively calculate the influence coefficients of the same text elements on the two legal clauses, and extract the difference features between the text elements; Step S600: A paragraph that is semantically associated with the first target paragraph is used as a second target paragraph. In the second target paragraph, the order of legal clauses recommended by the second target paragraph is adjusted according to the features of the elements in the legal clauses selected by the relevant personnel.

3. The text information intelligent analysis method based on a large language model according to claim 2, characterized in that: Step S200 includes: Step S201: Identify two text elements in the same paragraph in the text information, record them as the first text element and the second text element respectively, and form a tuple of any text element and the legal clause corresponding to the paragraph, wherein the tuple composed of the second text element and the corresponding legal clause is recorded as (A, C), wherein A represents the second text element, and C represents the legal clause corresponding to the same paragraph; Step S202: In the comparison database, the occurrence probability of the bigram (A, C) is calculated, and the occurrence probability pE of the first text element under the condition that the bigram (A, C) appears is calculated, pE=P(E|(A, C)), where E represents the first text element, and P() represents the probability calculation formula; Step S203: Obtain a tuple (E, C) consisting of the first text element and the legal clause corresponding to the first target segment, and calculate the probability pA of the second text element appearing under the condition that the tuple (E, C) appears, pA=P(A|(E, C)); Step S204: record pE as the influence coefficient of the first text element on the legal clause corresponding to the same paragraph, and record pA as the influence coefficient of the second text element on the legal clause corresponding to the same paragraph.

4. The text information intelligent analysis method based on a large language model according to claim 3 is characterized by: Step S300 includes: Step S301: obtaining a number of legal clauses matched by the first target paragraph, and obtaining the call probability of each legal clause; Step S302: Arrange the call probabilities of the legal clauses and the first target paragraph from high to low, and obtain an initial recommendation sequence of the legal clauses corresponding to the first target paragraph; Step S303: obtaining the legal clause with the highest call probability in the initial recommendation sequence and recording it as the first target clause of the first target paragraph, and obtaining the call probability of the legal clause in the first target paragraph and recording it as α.

5. The text information intelligent analysis method based on a large language model according to claim 4 is characterized in that: Step S400 includes: Step S401: In the second adjustment record, the legal clause selected by the relevant personnel is recorded as the second target clause of the first target paragraph; Step S402: In the first target paragraph, the calling probability of obtaining the second target clause is recorded as β, satisfying the condition β<α.

6. The text information intelligent analysis method based on a large language model according to claim 5, characterized in that: Step S500 includes: Step S501: Obtain the first text element and the second text element in the first target paragraph. Under the condition of the first target clause, the influence coefficient of the first text element is recorded as pE1, and the influence coefficient of the second text element is recorded as pA1. Under the condition of the second target clause, the influence coefficient of the first text element is recorded as pE2. The influence coefficient of the second text element is recorded as pA2. Step S502: when pE1>pA1 and pA2>pE2, it is the first adjustment mode. In the first adjustment mode, kd=pA2-pE2 is calculated; Step S503: When pE1>pA1, calculate k1=pE1-pA1, k2=pE2-pA2, wherein when k1<k2, it matches the second adjustment mode, and when k1>k2, it matches the third adjustment mode.

7. The text information intelligent analysis method based on a large language model according to claim 6, characterized in that: Step S600 includes: Step S601: Obtain an initial recommendation sequence of legal clauses corresponding to the second target paragraph, and establish an adjustment coefficient calculation function H of the legal clauses, H=μ+W, where μ represents the call probability of the legal clauses and the second target paragraph, and W represents the adjustment weight; Step S602: Obtain the influence coefficient of the first text element in the second target paragraph, record it as pE3, and the influence coefficient of the second text element, record it as pA3; Step S603: acquiring corresponding legal clauses in the initial recommendation sequence of the second target paragraph according to the matching conditions of the first adjustment mode, the second adjustment mode, and the third adjustment mode respectively; Step S604: When pE1>pA1, pA2>pE2 and pA3>pE3, match the first adjustment mode, obtain the legal terms that meet the condition pA3>pE3 in the legal terms corresponding to the second target paragraph, calculate ke=pA3-pE3, obtain the call probability μ1 of the legal terms and the second target paragraph, and calculate the adjustment coefficient H1 under the first adjustment mode, H1=μ1+(ke / kd)×(α-β); When pE1>pA1, pA3<pE3, and k1<k2, match the second adjustment mode, obtain the legal terms that meet the condition of pA3<pE3 in the legal terms corresponding to the second target paragraph, obtain the call probability μ2 of the legal terms and the second target paragraph, and calculate the adjustment coefficient H2 under the second matching mode, H2=μ2+(k3 / k2)×(α-β), where k3=pE3-pA3; When pE1>pA1, pA3<pE3, and k1>k2, match the third adjustment mode, obtain the legal terms that meet the condition of pA3<pE3 in the legal terms corresponding to the second target paragraph, obtain the call probability μ3 of the legal terms and the third target paragraph, and calculate the adjustment coefficient H3 under the third adjustment mode, H3=μ3+(k2 / k3)×(α-β), where k3=pE3-pA3; Step S605: In the initial recommendation sequence of the legal terms corresponding to the second target paragraph, the adjustment coefficient is used to replace the call probability of the corresponding legal terms, and the values ​​in the initial recommendation sequence are rearranged from high to low to obtain a recommended update sequence. The legal terms in the recommended update sequence are arranged from high to low according to the call probability or adjustment coefficient corresponding to the legal terms, and the legal terms are recommended to relevant personnel.

8. A text information intelligent analysis system based on a large language model, used to execute a text information intelligent analysis method based on a large language model as described in any one of claims 1 to 7, characterized in that: The system includes the following modules: a comparison database management module, an influence coefficient calculation module, an association matching module, a target adjustment management module, a difference calculation module and an adjustment management module, wherein the comparison database management module is used to manage the comparison database, the influence coefficient calculation module is used to calculate the influence coefficient of the text element, the association matching module is used to obtain the legal terms matched by each paragraph, the target adjustment management module is used to record the target adjustment, the difference calculation module is used to calculate the difference degree of the text elements corresponding to the legal terms before and after the target adjustment, and the adjustment management module is used to adjust the order of the legal terms matched by the second target paragraph.

9. The text information intelligent analysis system based on a large language model according to claim 8, characterized in that: The influence coefficient calculation module includes: a text element acquisition unit, a bigram management unit and an influence coefficient calculation unit, wherein the text element acquisition unit is used to acquire text elements in text information, the bigram management unit is used to manage bigrams consisting of the first text element or the second text element and the corresponding legal clause, and the first influence coefficient calculation unit is used to calculate the influence coefficient of the second text element of the first text element in the first target paragraph; The association matching module includes: an association matching model management unit and a first sorting unit, wherein the association matching model management unit is used to manage the legal terms calling model, and the first sorting unit is used to obtain the initial recommended sequence of paragraphs in the target text.

10. The text information intelligent analysis system based on a large language model according to claim 8, characterized in that: The target adjustment management module includes: a target adjustment acquisition unit and a call probability comparison unit, wherein the target adjustment acquisition unit is used to acquire the target adjustment, and the call probability comparison unit is used to compare the first target clause and the second target clause of the first target paragraph; The difference calculation module includes: an impact difference calculation unit and an adjustment mode acquisition unit, wherein the impact difference calculation unit is used to calculate the difference of the impact coefficient under different legal terms in the first target paragraph, and the adjustment mode acquisition unit is used to acquire the adjustment mode in the second target paragraph according to the difference of the impact coefficient; The adjustment management module includes: a second influence coefficient calculation unit, a pattern matching unit, an adjustment coefficient calculation unit and a second sorting unit, wherein the second influence coefficient calculation unit is used to calculate the influence coefficient of the second text element of the first text element in the second target paragraph, the pattern matching unit is used to match the adjustment pattern corresponding to each legal clause in the legal clause corresponding to the second target paragraph according to the adjustment pattern, the adjustment coefficient calculation unit is used to calculate the adjustment coefficient under each matching pattern using the adjustment coefficient calculation function, and the second sorting unit is used to update the recommended sequence of the legal provisions.

Images