CN115687563A

CN115687563A - Interpretable intelligent judgment method and device, electronic equipment and storage medium

Info

Publication number: CN115687563A
Application number: CN202210908771.5A
Authority: CN
Inventors: 陈浩; 于雪莉; 刘智静; 王华卿; 韩文娟; 弓琰
Original assignee: Beijing General Artificial Intelligence Research Institute
Current assignee: Beijing General Artificial Intelligence Research Institute
Priority date: 2022-07-29
Filing date: 2022-07-29
Publication date: 2023-02-03

Abstract

The invention provides an interpretable intelligent judgment method, an interpretable intelligent judgment device, electronic equipment and a storage medium. Through the mode, the structured map corresponding to the case description file can be obtained, equivalently, the events included in the case description file are disassembled, the events can be displayed in the structured map form, the updated structured map can be obtained based on the preset intelligent judgment and the preset map, the judgment result is obtained, the intelligent judgment process is clear and transparent, and the effects that the intelligent judgment can be explained and traced can be achieved.

Description

Interpretable intelligent judgment method and device, electronic equipment and storage medium

Technical Field

The present invention relates to the field of data processing technologies, and in particular, to an interpretable intelligent decision method, an interpretable intelligent decision device, an electronic device, and a storage medium.

Background

In recent years, with the increasing number of cases accepted by courts in various jurisdictions, the workload of judges is also increasing, and even if the officers have very high professional qualities under such high working pressure, the employment of mistakes such as erroneous judgment is inevitable.

At present, the intelligent decision mode is: based on the neural network model, information of all levels of case-by-case labels to which the case belongs is identified, for example, the case-by-case label is a divorce. On the basis, for example, the emotional breaking dispute under the divorce dispute is identified as the first stage case, and the judgment information corresponding to the label information of each stage is returned based on the label information of each identified stage case. Therefore, judges can be assisted, the working pressure of the judges is relieved, the working efficiency of the court is greatly improved, and misjudgments are reduced to the maximum extent.

However, the current intelligent judgment mode needs a large amount of training data to obtain the label of each case at each level, and the whole operation is a black box operation, so that the reason why the case belongs to the label cannot be explained. Therefore, when a judge makes a judgment by using the current intelligent judgment mode, the judge can only rely on the result, the reason and the key factors cannot be traced back, once the result is unstable and unreliable, the judge cannot correct the intermediate result, and only can manually judge from the beginning, but the workload is increased.

Disclosure of Invention

The invention provides an interpretable intelligent judgment method, an interpretable intelligent judgment device, electronic equipment and a storage medium, which are used for solving the defect that the judging result of intelligent judgment in the prior art cannot be traced, so that the intelligent judgment process is clear and transparent, and the interpretable and traceable effects of intelligent judgment can be achieved.

The invention provides an interpretable intelligent judgment method, which comprises the following steps:

acquiring a case description file;

extracting information of the case description file to obtain an extraction result;

constructing a structured graph based on the extraction result;

updating nodes of the structural map based on a preset intelligent judgment and OR graph and the structural map to obtain an updated structural map;

and generating a judgment result based on the node information of the updated structured graph.

Optionally, the step of extracting information from the case description file to obtain an extraction result includes:

preprocessing the case description file to obtain a preprocessed file;

performing sentence-level information extraction based on the preprocessed files to obtain sentence-level extraction results, and processing the sentence-level extraction results to obtain processing results, wherein the processing results comprise processing results corresponding to a plurality of event types;

and simplifying the processing result to obtain an extraction result.

Optionally, the preprocessed file includes a plurality of sentences to be processed;

the step of performing sentence-level information extraction based on the preprocessed files to obtain sentence-level extraction results and processing the sentence-level extraction results to obtain processing results comprises the following steps:

aiming at each sentence to be processed included in the preprocessed file, performing word segmentation processing on the sentence to be processed to obtain a word segmentation result;

performing part-of-speech tagging on the word segmentation result to obtain a part-of-speech tagging result;

carrying out entity naming identification on the word segmentation result to obtain an entity naming identification result;

performing semantic role labeling on the sentence to be processed to obtain a semantic role labeling result;

extracting a semantic dependency graph from the sentence to be processed to obtain a semantic dependency graph;

and obtaining a processing result based on a preset event structured summary, the part of speech tagging result, the entity naming identification result, the semantic role tagging result and the semantic dependency graph.

Optionally, the intelligent decision and or graph includes a plurality of event information and a plurality of nodes corresponding to the episode identification, and the event information and the nodes have a corresponding relationship;

the step of updating the nodes of the structured graph based on the pre-established intelligent judgment and OR graph and the structured graph to obtain an updated structured graph comprises the following steps:

identifying event information corresponding to each node aiming at each node corresponding to the plot identification included in the intelligent judgment and OR graph based on the structured graph to obtain an identification result;

and updating the identification result to the structural map to obtain an updated structural map.

Optionally, after the step of obtaining the case description file, the method further includes:

inputting the case description file into a pre-trained structure diagram to generate a model, and obtaining a current case diagram;

calculating the similarity between the current case diagram and a case diagram acquired in advance, wherein the case diagram acquired in advance is generated based on cases in a preset case library;

and taking the case with the similarity meeting the preset condition as a case to be recommended and pushing the case to be recommended.

Optionally, the case description file includes a fact information file and a general information file;

the step of constructing a structured atlas based on the extraction result comprises the following steps:

extracting preset element information corresponding to different event types based on the extraction result;

extracting preset general information based on the extraction result;

and forming a structural map based on the preset element information and the preset general information.

The present invention also provides an interpretable intelligent decision device, the device comprising:

the acquisition module is used for acquiring case description files;

the extraction module is used for extracting information of the case description file to obtain an extraction result;

the construction module is used for constructing a structural map based on the extraction result;

the updating module is used for updating the nodes of the structural map based on a preset intelligent judgment and OR graph and the structural map to obtain an updated structural map;

and the generating module is used for generating a judgment result based on the node information of the updated structured graph.

The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of an interpretable intelligent decision method as claimed in any one of the preceding claims when executing the program.

The invention also provides a non-transitory computer-readable storage medium having stored thereon a computer program which, when executed by a processor, carries out the steps of an interpretable intelligent decision method as claimed in any one of the preceding claims.

The invention also provides a computer program product comprising a computer program which, when executed by a processor, carries out the steps of an interpretable intelligent decision method as claimed in any one of the preceding claims.

According to the interpretable intelligent judgment method, the interpretable intelligent judgment device, the electronic equipment and the storage medium, the case description file is obtained, information of the case description file is extracted to obtain an extraction result, the structural map is constructed based on the extraction result, the nodes of the structural map are updated based on the preset intelligent judgment and OR graph and the structural map to obtain an updated structural map, and the judgment result is generated based on the node information of the updated structural map. Through the mode, the structured map corresponding to the case description file can be obtained, equivalently, the events included in the case description file are disassembled, the events can be displayed in the structured map form, the updated structured map can be obtained based on the preset intelligent judgment and the preset map, the judgment result is obtained, the intelligent judgment process is clear and transparent, and the effects that the intelligent judgment can be explained and traced can be achieved.

Drawings

In order to more clearly illustrate the technical solutions of the present invention or the prior art, the drawings needed for the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings according to the drawings without creative efforts.

FIG. 1 is a schematic flow chart of an interpretable intelligent decision method according to the present invention;

FIG. 2 is a second schematic flowchart of an interpretable intelligent decision method according to the present invention;

FIG. 3 is a schematic diagram of an intelligent decision AND-OR graph provided by the present invention;

fig. 4 is a third schematic flowchart of an interpretable intelligent decision method provided by the present invention;

FIG. 5 is a schematic representation of a structural map provided by the present invention;

FIG. 6 is a schematic diagram of an interpretable intelligent decision device according to the present invention;

fig. 7 is a schematic structural diagram of an electronic device provided by the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is obvious that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In order to make the intelligent judgment process clear and transparent, the effects of interpretability and traceability of the intelligent judgment can be achieved. The invention discloses an interpretable intelligent decision method, an interpretable intelligent decision device, electronic equipment, a non-transitory computer readable storage medium and a computer program product.

As shown in fig. 1, an interpretable intelligent decision method, the method comprising:

s101, acquiring a case description file.

When the case needs to be intelligently judged, the case description file of the case can be obtained from the case database, and the case description file of the case can also be input or imported to the electronic equipment through the user, so that the case description file can be obtained reasonably. The case database can comprise a plurality of case databases, and the case types corresponding to the case databases are different.

And S102, extracting information of the case description file to obtain an extraction result.

After the case description file is obtained, information extraction can be performed on the case description file to obtain an extraction result, wherein the extraction result can include results corresponding to different event types.

In one embodiment, the case description file can be preprocessed, sentence-level extracted, extracted and then processed, chapter-level extracted and the like, and the extraction of word segmentation, part-of-speech tagging, named entity recognition, semantic role tagging, semantic dependency graph extraction and the like is completed based on a natural language processing technology, so that an extraction result can be obtained.

The Named Entity Recognition (NER) is an Entity with a specific meaning in a Recognition text, and mainly includes a name of a person, a name of a place, a name of an organization, a proper noun, and the like. Semantic role labeling may label words and/or semantic roles that words serve in a sentence. Semantic dependency graphs are graphs that can characterize semantic associations between words and/or words in a sentence.

S103, constructing a structural map based on the extraction result.

After the extraction results are obtained, a structured atlas may be constructed based on the extraction results. In one embodiment, the fact information corresponding to each event type included in the scenario description file and the general information included in the scenario description file may be extracted based on the extraction result, so that a structured graph capable of covering the fact information corresponding to each event included in the scenario description file and the general information may be formed.

In one embodiment, the structured graph may use the participator as a node center, and further extend fact information corresponding to each event type participator, personal information of the participator, general information corresponding to the participator, and the like to the periphery.

As another embodiment, the structured graph may use the case name as a node center, and further extend, to the periphery, fact information corresponding to each event type corresponding to the case name, personal information of the participant, general information corresponding to the participant, and the like. This is all reasonable.

And S104, updating the nodes of the structural map based on a preset intelligent judgment and OR graph and the structural map to obtain an updated structural map.

After the structured map is obtained, the nodes of the structured map can be updated based on the preset intelligent judgment and OR graph and the structured map, so that the updated structured map is obtained. The intelligent judgment and or map can be made based on legal provisions of case types corresponding to case description files, and can comprise case identification.

In one embodiment, the situation identification included in the intelligent judgment and the graph can be identified based on the intelligent judgment generated by the legal provision and the structured graph capable of covering the fact information and the general information corresponding to each event type included in the case description file, so as to obtain the identification result, and therefore, the node of the structured graph can be updated based on the identification result, and the updated structured graph is obtained.

And S105, generating a judgment result based on the node information of the updated structured graph.

After the updated structured graph is obtained, a decision result may be generated based on the node information of the updated structured graph. Wherein the updated structured graph includes a plurality of nodes. In one embodiment, the node information for updating the structured graph may be input to a decision model trained in advance, so that a decision result may be obtained.

The decision model trained in advance can be a decision tree and a linear regression model trained according to the decision result of the historical case.

Therefore, the interpretable intelligent judgment method provided by the invention has the advantages that the case description file is obtained, the information of the case description file is extracted to obtain an extraction result, the structured graph is constructed based on the extraction result, the nodes of the structured graph are updated based on the preset intelligent judgment and OR graph and the structured graph to obtain an updated structured graph, and the judgment result is generated based on the node information of the updated structured graph. Through the mode, the structural map corresponding to the case description file can be obtained, which is equivalent to the disassembly of the events included in the case description file, and the events can be displayed in the form of the structural map, rather than the simple listing of the characteristics corresponding to the case, so that the interpretability of the case can be increased. And an updated structured map can be obtained based on a preset intelligent judgment and OR diagram, so that a judgment result is obtained, the intelligent judgment process is clear and transparent, and the effects of interpretability and traceability of the intelligent judgment can be achieved. And different decisions can be made for different situations.

As an embodiment of the present invention, the decision result may be updated to the update structured graph, and the node referred to by the decision result may also be labeled, so that interpretability and traceability may be further increased.

All information in the case is in the map, and each step of plot identification and judgment basis for intelligent judgment can also be displayed in the map, so that the interpretability and the traceability are further increased.

As an embodiment of the present invention, as shown in fig. 2, the step of extracting information from the case description file to obtain an extraction result may include:

s201, preprocessing the case description file to obtain a preprocessed file.

After the case description file is obtained, the case description file can be preprocessed to obtain a preprocessed file, wherein the preprocessing mode can include sentence division processing, parenthesis analysis, name attribute extraction, name processing and the like.

In order to more conveniently understand the interpretable intelligent judgment method disclosed by the invention, the preprocessing mode is introduced as follows:

the case description part included in the case description file is a text with a paragraph as a unit, and the content corresponding to a paragraph of text may include fact information corresponding to a plurality of event types. In order to obtain results of different event types, the words in paragraph units included in the case description file may be sentence-divided.

In one embodiment, the clauses may be divided according to punctuation marks, and the punctuation marks may be comma, period, semicolon, and other punctuation marks, which are not limited herein. And taking the case description file subjected to sentence division processing as a sentence division processing result.

The case description file can comprise a plurality of types of brackets, the case description file can be subjected to bracket analysis, the brackets in the case description file can be identified, and therefore the content near the brackets and the content in the brackets are obtained, and therefore the content near the brackets and the content in the brackets are processed correspondingly based on the type of the brackets.

For example, the content in the parentheses may be an explanation of a part of the content before the parentheses, and after the parentheses are identified, the nouns in the parentheses and the nouns before the parentheses may be acquired, thereby determining whether the nouns in the parentheses and the nouns before the parentheses are the academic names. Thus, the nouns which are the academic names are reserved, so that the extraction amount can be reduced under the condition of sentence-level extraction in the following.

After the case description file is obtained, the case description file can be identified to obtain the name of the person included in the case description file. The case description file may include a plurality of person names, and the attributes corresponding to each person name may be extracted to determine the corresponding participant. The attribute corresponding to the name of the person may include the first participant, the second participant, the third participant, and the like. In one embodiment, after the name included in the case description file is obtained, the name of the first participant may be searched for. Thereby completing the name attribute extraction.

In the case description file, the name of a person may be abbreviated, for example, zhang san may be named zhang. Therefore, for the convenience of subsequent intelligent judgment, the case description file can be identified to obtain the name of the person included in the case description file, and the name is replaced by a full name.

The execution sequence of the preprocessing mode can be fine clause, bracket analysis, name attribute extraction and name processing which are sequentially executed. Namely, after the sentence division processing result is obtained, bracket analysis, name attribute extraction and name processing are carried out.

Fine clauses, parenthesis analysis, name attribute extraction, and name processing may also be performed simultaneously. And fine sentence division can be performed after the parenthesis analysis, the name attribute extraction and the name processing are performed, and in this case, the sentence division processing result is the file to be processed. The adjustment can be performed according to actual needs, and is not specifically limited herein. This is all reasonable.

After the preprocessing is carried out, a file to be processed can be obtained, wherein the file to be processed comprises a plurality of sentences to be processed.

S202, sentence-level information extraction is carried out based on the preprocessed files, sentence-level extraction results are obtained and processed, and processing results are obtained.

After the preprocessed files are obtained, sentence-level information extraction can be performed based on the preprocessed files, and sentence-level extraction results are obtained and processed to obtain processing results. The processing result comprises processing results corresponding to a plurality of event types.

In an embodiment, each to-be-processed sentence in the preprocessed file may be subjected to word segmentation, part-of-speech tagging, named entity recognition, semantic role tagging, and extraction of a semantic dependency graph, so as to obtain a sentence-level extraction result, and the sentence-level extraction result is classified based on a preset event type and a preset trigger word, so as to obtain processing results, that is, processing results, corresponding to a plurality of event types.

Wherein, based on different types of cases, different event types and different trigger words can be preset.

S203, simplifying the processing result to obtain an extraction result.

After the processing result is obtained, the structured graph can be more quickly constructed in the follow-up process, and the processing result can be simplified, so that chapter-level extraction results can be obtained, wherein the extraction results comprise extraction results corresponding to a plurality of event types, and the number of the event types in the extraction results is less than that of the event types in the processing result.

For example, the processing result may include processing results corresponding to event a, event B, event C, and event D, where there may be overlap between the results corresponding to event a, event B, and event D, and therefore, simplified processing may be performed on the processing result to obtain an extraction result, and the extraction result may include extraction results corresponding to event a and event C.

In an embodiment, chapter-level extraction may be performed on the processing results, and the chapter-level processing results corresponding to each paragraph are extracted through a preset chapter schema (synopsis), so that event deduplication, event merging, and event completion are performed on the processing results corresponding to each paragraph. And the time information and the unit information in the processing result can be standardized.

Therefore, the interpretable intelligent judgment method disclosed by the invention can be used for preprocessing the case description file to obtain a preprocessed file, extracting sentence-level information based on the preprocessed file to obtain a sentence-level extraction result, processing the sentence-level extraction result to obtain a processing result, and simplifying the processing result to obtain an extraction result. The method is equivalent to the pretreatment, sentence-level extraction, reprocessing after extraction, chapter-level extraction and other treatments of case description files, so that extraction results are obtained, and the accuracy and efficiency of constructing the structured map can be improved.

As an embodiment of the present invention, the preprocessed file may include a plurality of sentences to be processed. And each sentence to be processed is subjected to bracket analysis, name attribute extraction and name processing.

The step of performing sentence-level information extraction based on the preprocessed file to obtain a sentence-level extraction result, and processing the sentence-level extraction result to obtain a processing result may include:

and aiming at each sentence to be processed included in the preprocessed file, performing word segmentation on the sentence to be processed to obtain a word segmentation result. And performing part-of-speech tagging on the word segmentation result to obtain a part-of-speech tagging result. Wherein, the part-of-speech tagging is a part-of-speech tag of the language unit tagging.

For example, word segmentation and part-of-speech tagging may be performed based on pre-designed natural language processing software, and after the word segmentation result is obtained, the word and/or part-of-speech of the word included in the word segmentation result may be tagged.

In one embodiment, in order to further improve the accuracy of part-of-speech tagging, after the part-of-speech tagging result is obtained, the part-of-speech tagging result may be adjusted, so that the structured graph can be generated more quickly in the following.

For example, when natural language processing software is used for word segmentation and part-of-speech tagging, a numerical value and a unit corresponding to the numerical value are disassembled and tagged, which is inconvenient for subsequently constructing a structural map. Therefore, after the part-of-speech tagging result is obtained, the numerical values in the analysis result and the units corresponding to the numerical values can be combined, so that the structural map can be constructed subsequently.

As an embodiment, since the to-be-processed sentence may include the indication pronouns, when the structured graph is subsequently constructed, the information may be ambiguous, and therefore, after the word segmentation result is obtained, the information corresponding to the indication pronouns in the to-be-processed sentence may be complemented at a sentence level. So that subsequently informative structured maps can be constructed.

And carrying out entity naming recognition on the word segmentation result to obtain an entity naming recognition result.

In one embodiment, an external NER dictionary may be introduced to better perform word segmentation, and after the word segmentation result is obtained, the word segmentation result may be input to a pre-trained entity name recognition model, so that an entity name recognition result may be obtained.

And performing semantic role labeling on the sentence to be processed to obtain a semantic role labeling result. And extracting the semantic dependency graph of the sentence to be processed to obtain the semantic dependency graph.

In one embodiment, each sentence to be processed may be input to a pre-trained semantic character tagging model to obtain a semantic character tagging result. And inputting each sentence to be processed into a pre-trained semantic dependency graph extraction model to obtain a semantic dependency graph.

The preset event structured summary may include different event types and trigger words and event elements corresponding to the event types. For example, the event types may include event type 1, event type 2 … event type n. The trigger word corresponding to the event type 1 may include a trigger word 1 and a trigger word 2 … trigger word m, and the event element corresponding to the event type 1 may include an event element 1 and an event element 2 … event element y.

After each sentence to be processed is respectively subjected to word segmentation, part-of-speech tagging, named entity recognition, semantic role tagging and extraction of a semantic dependency graph, aiming at each sentence to be processed, an event type corresponding to the sentence to be processed can be determined based on a preset event structured outline, and then a trigger word corresponding to the event type, a semantic role tagging result and/or an argument of the semantic dependency graph are determined from a part-of-speech tagging result, an entity naming recognition result, a semantic role tagging result and a semantic dependency graph corresponding to the sentence to be processed, so that a processing result corresponding to the sentence to be processed is obtained. And sequentially carrying out the processing on each sentence to be processed to obtain a processing result.

In an embodiment, multitask result fusion may be set, that is, result fusion is performed based on a semantic role labeling result, a semantic dependency graph, and a regular matching, where the regular matching is matching based on a preset rule. Thus, a more accurate processing result can be obtained.

As an embodiment of the present invention, before the step of updating the nodes of the structured graph based on the predetermined intelligent decision and or graph and the structured graph to obtain the updated structured graph, the method may further include:

and acquiring legal provisions corresponding to each case type, and formulating an intelligent judgment and OR graph corresponding to the case type, wherein the intelligent judgment and OR graph can cover each step required by judgment.

For example, case types may include case a, case b, etc., and are not specifically limited herein, and intelligent decisions and/or graphs formulated for case a may encompass participant selection, episode identification, result generation, and push.

In one embodiment, the intelligent decision and or graph may include nodes corresponding to each step required for decision, and event information corresponding to each node is different. For example, as shown in fig. 3, an And or graph, which may also be referred to as AOG (And/or graph), is created for the b-case, and a participant result node 301, and the participant result node 301 is connected to a participant selection node 302 And a result node 303. The result node 303 is connected to a first episode determination node 304, a result node 305 and a result push node 306. First episode determination node 304 is coupled to second episode determination node 307 and third episode determination node 314.

Second episode identification node 307 is coupled to first age node 308, first special identification node 309, participation attribute identification node 310, second special identification node 311, third special identification node 312, and standing identification node 313. The third episode identification node 314 is connected to a sub-episode identification node 315 and a second age node 316.

The result node 305 is connected to a first sub-result node 317, a second sub-result node 318, and a third sub-result node 322. The second sub-result node 318 is connected to a fourth sub-result node 319, a fifth sub-result node 320 and a sixth sub-result property 321. The third sub-result node 322 is connected to a fact-recognizing node 323, a range-determining node 324, a generate-suggested node 325, and a suggested-revise node 326.

Therefore, the interpretable intelligent judgment method disclosed by the invention can be used for generating the intelligent judgment and OR diagram corresponding to the case type according to the legal provisions corresponding to each case type, and can be used for conveniently obtaining accurate judgment results subsequently.

As an embodiment of the present invention, the intelligent decision and or graph may include a plurality of event information corresponding to the episode identification and a plurality of nodes, where the event information and the nodes have a corresponding relationship. Wherein the episode identification includes general episode identification and factual episode identification.

The step of updating the nodes of the structured graph based on the predetermined intelligent decision and or graph and the structured graph to obtain the updated structured graph may include:

and identifying the event information corresponding to each node aiming at each node corresponding to the plot identification included in the intelligent judgment and OR graph based on the structured graph to obtain an identification result.

After the structured graph corresponding to the case description file and the intelligent judgment and OR-graph corresponding to the case description file are obtained, the event information corresponding to the node can be identified according to each node corresponding to the case identification included in the intelligent judgment and OR-graph based on the structured graph, and the identification result is obtained.

In one embodiment, when event information corresponding to a certain node corresponding to the scene identification is identified, the node corresponding to the event information in the structured graph can be highlighted, and the identification of the event information corresponding to the node is completed, so that an identification result is obtained. Therefore, the intelligent judgment process can be further clear and transparent, and the effects of interpretability and traceability of the intelligent judgment can be achieved.

For example, the event information may be "considered age", and the node corresponding to the event information "considered age" may be scalar-structured with the node corresponding to the case time in the map and the node corresponding to the birth date of the participant, and the considered age of the participant may be calculated based on the case time and the birth date.

After the determination result is obtained, the determination result may be updated to the structured graph to obtain an updated structured graph, and in one embodiment, the determination result may be added to the structured graph as a new branch, so as to subsequently improve the accuracy of the generated decision result.

In an embodiment, after each node is identified, an identification result is obtained, and the identification result is updated to the structural map in real time until each node is identified, and the structural map is updated to obtain an updated structural map.

In another embodiment, after each node corresponding to the episode identification is identified, all the identification results of the nodes corresponding to the episode identification can be obtained, and then the identification results are updated to the structured graph to obtain the updated structured graph.

As an embodiment of the present invention, after the step of obtaining the case description file, the method may further include:

and inputting the case description file into a structural diagram generation model which is trained in advance to obtain a current case diagram.

After the case description file is obtained, the case description file can be input into a structural diagram generation model which is trained in advance, and a current case diagram is obtained. In one embodiment, the case description file is input into the structural diagram generation model which is trained in advance, and the structural diagram generation model which is trained in advance can learn the structural association among all nodes in the case description file, so that a fine-grained case diagram can be output, namely, the current case diagram can be obtained.

In another embodiment, the case basic situation corresponding to the updated structured graph can be input into the structural diagram generation model which is trained in advance, so that the current case graph can be obtained.

The pre-trained structure diagram generation model can be a diagram neural network model under an attention mechanism, and of course, other structure diagram generation models capable of generating fine-grained case diagrams can be selected according to actual needs, which is reasonable.

And calculating the similarity between the current case pattern and a case pattern acquired in advance.

After the current case diagram is acquired, the similarity between the structure of the current case diagram and a case diagram acquired in advance can be calculated, wherein the case diagram acquired in advance is generated based on cases in a preset case library.

In an embodiment, positive and negative samples may be sampled for a basic situation corresponding to each case in the case library according to a preset case similarity standard, and a case diagram representation learning is performed in a training mode of sequencing by using the positive and negative samples, so as to obtain a case diagram including each case in the case library, that is, the case diagram obtained in advance is a case diagram corresponding to all cases in the case library.

After the similarity between the current case diagram and the case diagram acquired in advance is calculated, cases with the similarity meeting preset conditions can be used as cases to be recommended and pushed. In an implementation manner, the current case pattern and the similarity corresponding to each case in the case pattern may be sorted, and then the case with the similarity at the top preset digit is taken as the case to be recommended, and the case to be recommended is pushed.

In another embodiment, a preset number of cases with similarity greater than a preset similarity threshold value are selected from the similarity between the current case pattern and each case in the case pattern, and are taken as cases to be recommended, and the cases to be recommended are pushed.

Therefore, the interpretable intelligent judgment method disclosed by the invention can be used for inputting the case description file into a structural diagram generation model which is trained in advance to obtain a current case diagram, and calculating the similarity between the current case diagram and a case diagram which is acquired in advance, wherein the case diagram which is acquired in advance is generated based on cases in a preset case library, and the cases with the similarity meeting preset conditions are used as cases to be recommended and pushed. Therefore, the case pushing based on the case description file can be realized, and judges can be more effectively assisted by judges.

As an implementation manner of the embodiment of the present invention, the case description file may include a fact information file and a general information file. The fact information file is a crime fact information file corresponding to the case, and the general information file is a general plot information file corresponding to the case.

As shown in fig. 4, the step of constructing a structural map based on the extraction result may include:

s401, extracting the preset element information corresponding to different event types based on the extraction result.

After the extraction result is obtained, preset element information corresponding to different event types may be extracted based on the extraction result, where the preset element information may be event element information (i.e., fact information) corresponding to each preset event type. It is reasonable that the event element information corresponding to each event type may be consistent or may not be consistent.

S402, extracting the preset general information based on the extraction result.

After the extraction result is obtained, the preset general information may be extracted based on the extraction result, where the preset general information may be preset general episode information.

The execution sequence of the steps S401 and S402 may be that the step S401 is executed first and then the step S402 is executed, or the step S402 is executed first and then the step S401 is executed, or the step S401 and the step S402 are executed simultaneously. This is all reasonable.

And S403, forming a structured map based on the preset element information and the preset general information.

After the preset element information and the preset general information are acquired, a structured map can be formed. In one embodiment, a first structural map may be generated based on preset element information, a second structural map may be generated based on preset general information, and further, a structural map may be formed based on the first structural map and the second structural map. The first structured graph is a case event graph capable of covering the occurrence sequence of case events and the key information of the case. The second structured graph is a general plot graph capable of covering the general plot information of the case.

In another embodiment, a structured graph can be formed based on the preset element information and the preset general information, and the structured graph can cover the case event graph of the case event occurrence sequence and the case key information and can also cover the general plot graph of the case general plot information.

As an implementation mode of the invention, after the updated structured map is obtained, the judgment experience, the flow, the industry consensus and the like of the law clauses and judges can be integrated to generate a more flexible and controllable judgment result. For example: the judge can make modifications and adjustments to any step in the decision process.

As an implementation mode of the invention, the updating of the structured map is expandable, more case sets can be added, and only the event types under the corresponding case sets need to be defined and extracted. And the updating of the structured graph can be maintained, for example, if legal provisions are modified, only the judgment operation logic needs to be modified on the corresponding nodes on the intelligent judgment and OR graph, and the traditional method such as a deep learning method is invalid immediately when the situation is met, and a new judgment model needs to be retrained after accumulating enough new data.

In order to facilitate understanding of the interpretable intelligent decision method disclosed by the present invention, an interpretable intelligent decision method of the present invention is described below with reference to examples:

for example, the case description file 1 is "b year, c day, d year" a year ", and the participator stone y sells 2 h to three in the f-way, g-way, 9-way, 1-building fire-fighting access of the e-area, and the g-cell of the local city at the price of the Renminbi x Yuan. The j components of the above products are all detected by identification, and the total weight is k grams. "

Information extraction is performed on the case description file 1 to obtain an extraction result, and further based on the extraction result, a structured graph can be obtained, wherein the structured graph can be connected with a case node 501, an inspection node 502, a selling node 503 and a stone y node 504 as shown in fig. 5.

The check node 502 is connected to a k node 505, a j node 506, and a permission node 507 for a year, b, month, c, d. The selling node 503 is connected with Zhang Sanjie point 508, node 509 at the fire fighting access of 1 building in the e-district f-way g-district 9 of the city, x-element node 510 and permission node 511 at b, month, c and d of a year.

The stone y node 504 is connected to the 1988/08/11 node 512, the stone y node 513, and the special plot node 514. Special case node 514 is connected to main node 515 and to tube node 516.

Furthermore, the nodes of the structural graph can be updated based on the AOG (And/or graph) as shown in fig. 3, so as to obtain an updated structural graph. Based on the node information of the updated structured graph, a decision result can be generated.

An interpretable intelligent decision device according to the present invention is described below, and an interpretable intelligent decision device described below and an interpretable intelligent decision method described above may be referred to in correspondence with each other.

As shown in fig. 6, the present invention discloses an interpretable intelligent decision device, which includes:

the obtaining module 610 is configured to obtain the case description file.

And the extraction module 620 is configured to extract information from the case description file to obtain an extraction result.

A construction module 630, configured to construct a structured graph based on the extraction result.

And the updating module 640 is used for updating the nodes of the structural map based on a preset intelligent judgment and OR graph and the structural map to obtain an updated structural map.

A generating module 650, configured to generate a decision result based on the node information of the updated structured graph.

As an embodiment of the present invention, the extraction module 620 may include:

and the preprocessing unit is used for preprocessing the case description file to obtain a preprocessed file.

And the sentence-level extraction unit is used for extracting sentence-level information based on the preprocessed file to obtain a sentence-level extraction result and processing the sentence-level extraction result to obtain a processing result.

Wherein the processing result comprises processing results corresponding to a plurality of event types.

And the simplification processing unit is used for simplifying the processing result to obtain an extraction result.

As an embodiment of the present invention, the preprocessed file includes a plurality of sentences to be processed.

The sentence level extraction unit may include:

and the word segmentation subunit is used for carrying out word segmentation on the sentences to be processed aiming at each sentence to be processed included in the preprocessed file to obtain word segmentation results.

And the first labeling subunit is used for performing part-of-speech labeling on the word segmentation result to obtain a part-of-speech labeling result.

And the recognition subunit is used for carrying out entity naming recognition on the word segmentation result to obtain an entity naming recognition result.

And the second labeling subunit is used for performing semantic role labeling on the sentence to be processed to obtain a semantic role labeling result.

And the extraction subunit is used for extracting the semantic dependency graph from the sentence to be processed to obtain the semantic dependency graph.

And the determining subunit is used for obtaining a processing result based on a preset event structured summary, the part-of-speech tagging result, the entity naming identification result, the semantic role tagging result and the semantic dependency graph.

As an embodiment of the present invention, the intelligent decision and or graph includes a plurality of event information corresponding to the episode identification and a plurality of nodes, and the event information and the nodes have a corresponding relationship.

The update module 640 may include:

and the identifying unit is used for identifying the event information corresponding to each node according to the structured graph aiming at each node corresponding to the plot identification included in the intelligent judgment and OR graph to obtain an identification result.

And the updating unit is used for updating the identification result to the structural map to obtain an updated structural map.

As an embodiment of the present invention, the apparatus may further include:

the input module is used for inputting the case description file into a pre-trained structure diagram generation model after acquiring the case description file to obtain a current case diagram;

and the calculating module is used for calculating the similarity between the current case picture and the case picture acquired in advance.

The case images acquired in advance are generated based on cases in a preset case library.

And the pushing module is used for taking the case with the similarity meeting the preset condition as a case to be recommended and pushing the case to be recommended.

As an embodiment of the present invention, the case description file includes a fact information file and a general information file.

The extraction module 620 may include:

and the first extraction unit is used for extracting the preset element information corresponding to different event types based on the extraction result.

And the second extraction unit is used for extracting the preset general information based on the extraction result.

And the forming unit is used for forming a structural map based on the preset element information and the preset general information.

Fig. 7 illustrates a physical structure diagram of an electronic device, and as shown in fig. 7, the electronic device may include: a processor (processor) 710, a communication Interface (Communications Interface) 720, a memory (memory) 730, and a communication bus 740, wherein the processor 710, the communication Interface 720, and the memory 730 communicate with each other via the communication bus 740. Processor 710 may invoke logic instructions in memory 730 to perform an interpretable intelligent decision method provided by the methods described above.

In addition, the logic instructions in the memory 730 can be implemented in the form of software functional units and stored in a computer readable storage medium when the software functional units are sold or used as independent products. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

In another aspect, the present invention also provides a computer program product comprising a computer program storable on a non-transitory computer readable storage medium, the computer program, when executed by a processor, being capable of executing an interpretable intelligent decision method provided by the above methods.

In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform an interpretable intelligent decision method provided by the above methods.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.

Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims

1. An interpretable intelligent decision method, the method comprising:

acquiring a case description file;

constructing a structured graph based on the extraction result;

updating nodes of the structural map based on a preset intelligent judgment and OR map and the structural map to obtain an updated structural map;

2. The interpretable intelligent decision method according to claim 1, wherein the step of extracting information from the case description file to obtain an extraction result comprises:

preprocessing the case description file to obtain a preprocessed file;

and simplifying the processing result to obtain an extraction result.

3. An interpretable intelligent decision method as claimed in claim 2, wherein the preprocessed file includes a plurality of sentences to be processed;

the step of performing sentence-level information extraction based on the preprocessed file to obtain a sentence-level extraction result and processing the sentence-level extraction result to obtain a processing result comprises the following steps:

for each sentence to be processed included in the preprocessed file, performing word segmentation on the sentence to be processed to obtain a word segmentation result;

carrying out entity naming recognition on the word segmentation result to obtain an entity naming recognition result;

extracting semantic dependency graphs of the sentences to be processed to obtain semantic dependency graphs;

and obtaining a processing result based on a preset event structured summary, the part-of-speech tagging result, the entity naming identification result, the semantic role tagging result and the semantic dependency graph.

4. The method according to claim 1, wherein the intelligent decision and or graph comprises a plurality of event information corresponding to episode identification and a plurality of nodes, and the event information and the nodes have corresponding relationship;

identifying the event information corresponding to each node aiming at each node corresponding to the plot identification included in the intelligent judgment and OR graph based on the structural graph to obtain an identification result;

5. An interpretable intelligent decision method according to any one of claims 1 to 4, wherein after the step of obtaining a case description file, the method further comprises:

6. An interpretable intelligent decision method according to claim 2 or 3, wherein the case description file includes a fact information file and a general information file;

extracting preset general information based on the extraction result;

7. An interpretable intelligent decision device, the device comprising:

the acquisition module is used for acquiring case description files;

8. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of an interpretable intelligent decision method according to any one of claims 1 to 6 when executing the computer program.

9. A non-transitory computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of an interpretable intelligent decision method as claimed in any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of an interpretable intelligent decision method as claimed in any one of claims 1 to 6 when executed by a processor.