Natural language based cross-platform workflow intelligent generation and interpretation system

Abstract

The application relates to the technical field of data processing, in particular to a cross-platform work flow intelligent generation and interpretation system based on natural language, which receives multi-modal language data through a language receiving module, verifies convergence of structured data through a language conversion module, and performs ambiguity analysis and semantic validity analysis on unstructured data, performs blurring and avoidance on private words in text logic through a privacy processing module, acquires a conflict-free first semantic intention through an intention understanding module, writes the conflict-free first semantic intention fingerprint into a block chain through a work flow generation module, judges the meeting degree of a user like amount and the meeting degree of a privacy score through a work flow interpretation module, and generates a work flow paython script cross-platform operation, so that the work flow can be efficiently realized on multiple platforms and be suitable for a work scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a cross-platform workflow intelligent generation and interpretation system based on natural language. BACKGROUND

[0002] At present, the digital transformation of enterprises has entered the deep water area. In the aspect of data security and privacy compliance, the traditional knowledge graph lacks an endogenous security mechanism in the "extraction-fusion-reasoning" whole link containing sensitive attributes. The static desensitization method destroys data correlation, while the graph calculation and reasoning under plaintext face privacy stealing risks such as member reasoning and model reverse. Secondly, in the natural language interaction layer, there is a significant gap between the system understanding of Chinese field slang and the user's real intention. The existing intention recognition model is like a "black box", its decision-making process cannot be verified and traced, and the generated intention cannot be bidirectionally checked with the underlying knowledge graph and data authority, resulting in a large number of "unexecutable" instructions and frequent user experience breakpoints.

[0003] The prior art still has the problems that the generated workflow cannot fit the work scene due to the lack of understanding of the semantics of the work environment and the standardization processing of slang, and the data cannot be traced and the privacy is insufficient due to the lack of use of blockchain and privacy processing. SUMMARY

[0004] Therefore, the present application provides a cross-platform workflow intelligent generation and interpretation system based on natural language to overcome the problems in the prior art that the generated workflow cannot fit the work scene due to the lack of understanding of the semantics of the work environment and the standardization processing of slang, and the data cannot be traced and the privacy is insufficient due to the lack of use of blockchain and privacy processing.

[0005] To achieve the above purpose, the present application provides a cross-platform workflow intelligent generation and interpretation system based on natural language, which comprises:

[0006] A language receiving module is configured to receive multi-modal language data, wherein the multi-modal language data comprises structured data and unstructured data.

[0007] A language conversion module is configured to convert the multi-modal language data into text logic, and to structure optimize the process of conversion into text logic, and to update the results of the process of structure optimization and the process of conversion into text logic.

[0008] A privacy processing module is configured to desensitize the text logic to obtain desensitized text logic, and to generate a target request fingerprint according to the desensitized text logic, and to standardize the target request fingerprint to obtain a semantic understanding result.

[0009] The intention understanding module is configured to output the first semantic intention according to the semantic understanding result, and acquire the non-conflict first semantic intention according to the first semantic intention;

[0010] The workflow generation module is configured to acquire the target semantic fingerprint according to the non-conflict first semantic intention, and generate the target workflow according to the target semantic fingerprint;

[0011] The workflow interpretation module is configured to perform work interpretation and process optimization according to the target workflow.

[0012] Further, the language conversion module comprises:

[0013] The logic conversion unit is configured to convert the multi-modal language data into literal logic;

[0014] The logic optimization unit is configured to perform convergence verification on the structured data to obtain a convergence verification result, and perform structural optimization on the process of converting the multi-modal language data into literal logic according to the convergence verification result;

[0015] The logic updating unit is configured to perform ambiguity analysis on the unstructured data to obtain an ambiguity analysis result, perform effectiveness analysis on the unstructured data according to the ambiguity analysis result to obtain an effectiveness analysis result, and perform result updating on the output process of the convergence verification result according to the effectiveness analysis result.

[0016] Further, the logic conversion unit converts the multi-modal language data into literal logic through a natural language processing technology;

[0017] The logic optimization unit performs convergence verification on the structured data, calculates the Pearson correlation coefficient P of the structured data and the previous structured data through Python, and -1

[0018] When -1

[0019] When 0

[0020] When the logic optimization unit performs structural optimization on the process of converting the multi-modal language data into literal logic, the domain database is used to assist the conversion of the multi-modal language data into literal logic.

[0021] Further, the logic updating unit performs ambiguity analysis on the unstructured data by natural language processing technology to obtain an ambiguity analysis result, the ambiguity analysis result including existence of a difference and non-existence of a difference, and performs semantic validity analysis on the unstructured data according to the ambiguity analysis result, wherein:

[0022] When the ambiguity analysis result is non-existence of a difference, the semantic validity analysis is not performed on the unstructured data;

[0023] When the ambiguity analysis result is existence of a difference, the semantic validity analysis is performed on the unstructured data by natural language processing technology to obtain an effectiveness analysis result, the effectiveness analysis result including valid semantics and invalid semantics;

[0024] The logic updating unit updates the output process of the convergence verification result according to the effectiveness analysis result, wherein:

[0025] When the effectiveness analysis result is invalid semantics, the output process of the convergence verification result is not updated, and the process of converting the literal logic is directly structurally optimized:

[0026] When the effectiveness analysis result is valid semantics, the output process of the convergence verification result is updated: the semantic features of the unstructured data are extracted by natural language processing technology to obtain unstructured semantic features, the unstructured semantic features are added to the pre-structured data, and the Pearson correlation coefficient P of the structured data and the pre-structured data is recalculated.

[0027] Further, the privacy processing module desensitizes the literal logic to obtain desensitized literal logic, and generates a target request fingerprint according to the desensitized literal logic by a hash function;

[0028] The privacy processing module performs black word standardization processing on the target request fingerprint, constructs a black word standardization processing model, and inputs the target request fingerprint into the black word standardization processing model to obtain a semantic understanding result output by the black word standardization processing model.

[0029] Further, the intent understanding module includes:

[0030] An intent generation unit is configured to obtain semantic confidence according to the semantic understanding result, update the process of black word standardization processing according to the semantic confidence, and output a first semantic intent;

[0031] An intent processing unit is configured to perform bidirectional verification on the first semantic intent by using a knowledge graph to obtain a bidirectional verification result, and obtain a conflict-free first semantic intent according to the bidirectional verification result.

[0032] Further, the intention generation unit obtains semantic confidence according to the semantic understanding result, and updates the process of the slang standardization processing according to the semantic confidence, takes the BERT allocation probability in the slang standardization processing model as the semantic confidence zc, compares the semantic confidence zc with a preset semantic confidence zc0, judges the compliance degree of the semantic confidence according to the comparison result, and updates the process of the slang standardization processing according to the judgment result, wherein:

[0033] When zc≥zc0, the intention generation unit determines that the compliance degree of the semantic confidence is up to standard, does not update the process of the slang standardization processing, and outputs the semantic understanding result as the first semantic intention;

[0034] When zc<zc0, the intention generation unit determines that the compliance degree of the semantic confidence is not up to standard, updates the process of the slang standardization processing: pushes the semantic understanding result to the user through the interactive interface for confirmation, obtains the confirmed semantic understanding result, and outputs the confirmed semantic understanding result as the first semantic intention.

[0035] Further, the intention processing unit performs bidirectional verification on the first semantic intention, constructs a semantic intention knowledge graph, and performs bidirectional verification on the first semantic intention according to the semantic intention knowledge graph to obtain a bidirectional verification result, wherein the bidirectional verification result includes verification pass and verification fail, and wherein:

[0036] When the bidirectional verification result is verification pass, the first semantic intention is subjected to semantic conflict resolution to obtain a conflict-free first semantic intention;

[0037] When the bidirectional verification result is verification fail, the first semantic intention is marked as missing and pushed to an administrator.

[0038] Further, the workflow generation module generates a workflow according to the conflict-free first semantic intention by a workflow generation method, wherein the workflow generation method comprises:

[0039] Step A01, obtaining a conflict-free first semantic intention fingerprint according to the conflict-free first semantic intention by a hash function, and writing the conflict-free first semantic intention fingerprint into a blockchain;

[0040] Step A02, generating a workflow Python script according to the conflict-free first semantic intention, and performing cross-platform container execution according to the workflow Python script to obtain a cross-platform execution result, wherein the cross-platform execution result includes execution success and execution failure;

[0041] Step A03, performing execution compensation on the process executed by the cross-platform container according to the cross-platform execution result, wherein:

[0042] When the cross-platform execution result is execution success, no execution compensation is performed on the process executed by the cross-platform container;

[0043] When the cross-platform execution result is execution failure, execution compensation is performed on the process executed by the cross-platform container: all business impacts and resource changes generated by the workflow Python script executed by the cross-platform container during the execution process.

[0044] Further, the workflow interpretation module performs work interpretation on the target workflow through a work interpretation method, and the work interpretation method comprises:

[0045] Step B01, generating a Chinese briefing graph through natural language processing technology according to the cross-platform execution result;

[0046] Step B02, pushing the Chinese briefing graph to the user end and the operation end, acquiring the user like amount Ds through the user end, and performing differential privacy testing on the Chinese briefing graph through the operation end to obtain a privacy score Ys;

[0047] Step B03, comparing the user like amount Ds with a preset user like amount Ds0, judging the compliance degree of the user like amount according to the comparison result, and outputting a first privacy interpretation decision according to the judgment result, wherein:

[0048] When Ds≥Ds0, it is determined that the compliance degree of the user like amount is up to standard, and the first privacy interpretation decision is output: new black words are mined from enterprise chat records and meeting minutes, and the new black words are added to the training library of the black word standardization processing model to retrain the black word standardization processing model;

[0049] When Ds<Ds0, it is determined that the compliance degree of the user like amount is not up to standard, and the first privacy interpretation decision is output: opinion feedback is pushed to the user through an interactive interface, and the opinion feedback result is sent to the operation and maintenance personnel;

[0050] Step B04, comparing the privacy score Ys with a preset privacy score Ys0, setting the preset privacy score Ys0=0.85, judging the compliance degree of the privacy score according to the comparison result, and outputting a second privacy interpretation decision according to the judgment result, wherein:

[0051] When Ys≥Ys0, it is determined that the compliance degree of the privacy score is up to standard, and the second privacy interpretation decision is output: the conflict-free first semantic intention is marked as an old intention, the availability of the historical old intention is automatically checked, and the user is reminded to update the system according to the availability of the historical old intention;

[0052] When Ys < Ys0, it is determined that the compliance degree of the privacy score is unqualified, and the second privacy explanation decision is output: the operation end supplements the entity node in the semantic intention knowledge graph to obtain a supplemented semantic intention knowledge graph, and re-performs bidirectional verification on the first semantic intention according to the supplemented semantic intention knowledge graph.

[0053] Compared with the prior art, the system receives multi-modal language data through the language receiving module, and the system also performs convergence verification on structured data through the language conversion module, and performs ambiguity analysis and semantic validity analysis on unstructured data, so as to add field database auxiliary judgment in the process of converting multi-modal language data into literal logic, thereby improving the accuracy of literal logic, the system also performs fuzzing and avoidance on the privacy vocabulary in the literal logic through the privacy processing module, thereby improving data security, the privacy processing module also makes the semantic understanding result fit the use scene through the slang standardization processing, thereby improving the accuracy and adaptability of the work flow generation, the system also acquires the conflict-free first semantic intention through the intention understanding module, and writes the conflict-free first semantic intention fingerprint into the block chain through the work flow generation module, permanent saving while being traceable, at the same time, the work flow generation module generates the work flow Python script and performs cross-platform container execution to verify the feasibility of cross-platform execution of the work flow Python script, the system also judges the compliance degree of the user like amount and the compliance degree of the privacy score through the work flow explanation module, so as to guarantee the work flow matching and privacy of the work flow Python script generation, and outputs the first privacy explanation decision and the second privacy explanation decision to further improve the privacy protection of the work flow generation. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 It is a structural schematic diagram of the natural language-based cross-platform work flow intelligent generation and explanation system of the embodiment;

[0055] Figure 2 It is a flowchart of the work flow generation method of the embodiment;

[0056] Figure 3 It is a flowchart of the work flow explanation method of the embodiment. DETAILED DESCRIPTION

[0057] In order to make the purpose and advantages of the present application clearer and more apparent, the present application will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the present application.

[0058] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art will understand that the embodiments are only used to explain the technical principles of the present application and are not intended to limit the protection scope of the present application.

[0059] It should be noted that, in the description of the present application, the terms indicating the direction or positional relationship of "upper", "lower", "left", "right", "inner", "outer" and the like are based on the direction or positional relationship shown in the drawings, which is only for the convenience of description, and does not indicate or imply that the device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0060] In addition, it should be noted that, in the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through intermediate medium, or internal communication of two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0061] Please refer to Figure 1 As shown in the figure, it is a structural schematic diagram of the natural language cross-platform workflow intelligent generation and interpretation system based on the embodiment, which comprises:

[0062] The language receiving module is used to receive multi-modal language data, which includes structured data and unstructured data;

[0063] The language conversion module is used to convert the multi-modal language data into text logic, and to structure optimize the process of converting into text logic, and to update the results of the process of structure optimization and the process of converting into text logic, and is connected with the language receiving module;

[0064] The privacy processing module is used to desensitize the text logic to obtain desensitized text logic, and to generate a target request fingerprint according to the desensitized text logic, and to standardize the target request fingerprint according to the black word, to obtain the semantic understanding result, and is connected with the language conversion module;

[0065] The intent understanding module is used to output the first semantic intent according to the semantic understanding result, and to acquire the non-conflict first semantic intent according to the first semantic intent, and is connected with the privacy processing module;

[0066] a workflow generation module configured to obtain a target semantic fingerprint according to the conflict-free first semantic intention, and generate a target workflow according to the target semantic fingerprint, the workflow generation module being connected to the intention understanding module;

[0067] a workflow interpretation module configured to perform work interpretation and process optimization according to the target workflow, the workflow interpretation module being connected to the workflow generation module.

[0068] Specifically, the natural language-based cross-platform workflow intelligent generation and interpretation system is applied to office equipment of large enterprises and groups. The system receives multi-modal language data, obtains conflict-free first semantic intentions according to the multi-modal language data, and generates workflow Python scripts for cross-platform running according to the conflict-free first semantic intentions, so as to realize efficient and work-scene-conforming workflows on multiple platforms. The system receives multi-modal language data through a language receiving module. The system also verifies the convergence of structured data through a language conversion module, and performs ambiguity analysis and semantic validity analysis on unstructured data, so as to add domain database auxiliary judgment in the process of converting multi-modal language data into text logic, thereby improving the accuracy of text logic. The system also blurs and avoids private words in text logic through a privacy processing module, thereby improving data security. The privacy processing module also processes slang standardization to make semantic understanding results fit the use scenario, thereby improving the accuracy and adaptability of workflow generation. The system obtains conflict-free first semantic intentions through an intention understanding module, and writes the conflict-free first semantic intention fingerprints into a blockchain through a workflow generation module, which is permanently saved while being traceable. Meanwhile, the workflow generation module generates workflow Python scripts and performs cross-platform container execution to verify the feasibility of cross-platform execution of the workflow Python scripts. The system also judges the compliance degree of the user's like amount and the compliance degree of the privacy score through a workflow interpretation module, so as to ensure the workflow matching and privacy of the generated workflow Python scripts, and outputs the first privacy interpretation decision and the second privacy interpretation decision to further improve the privacy protection of workflow generation.

[0069] Specifically, the language receiving module receives multi-modal language data, which includes structured data and unstructured data.

[0070] Specifically, the multi-modal language data refers to multi-modal data in the form of text, voice and pictures representing business requirements input by a user through a multi-platform input interface, the structured data refers to data with a predefined model, fixed format and easy to be directly processed by a computer program, such as date and time and digital amount, and the unstructured data refers to data without a predefined format and requiring analysis by a computer program to understand the content, such as natural language description and sentiment orientation.

[0071] Specifically, the language conversion module comprises:

[0072] a logic conversion unit configured to convert the multi-modal language data into literal logic;

[0073] a logic optimization unit configured to perform convergence verification on the structured data to obtain a convergence verification result, and perform structural optimization on the process of converting into literal logic according to the convergence verification result, the logic optimization unit being connected with the logic conversion unit;

[0074] a logic updating unit configured to perform ambiguity analysis on the unstructured data to obtain an ambiguity analysis result, perform effectiveness analysis on the unstructured data according to the ambiguity analysis result to obtain an effectiveness analysis result, and perform result updating on the output process of the convergence verification result according to the effectiveness analysis result, the logic updating unit being connected with the logic optimization unit.

[0075] Specifically, the logic conversion unit converts the multi-modal language data into literal logic through natural language processing technology.

[0076] Specifically, the logic optimization unit performs convergence verification on the structured data, calculates the Pearson correlation coefficient P of the structured data and the previous structured data through Python, and -1

[0077] When -1

[0078] When 0

[0079] When 0

[0080] Specifically, the literal logic refers to a machine-readable semantic framework with explicit structural relationship and semantic meaning extracted from the multi-modal language data, the natural language processing technology refers to an existing technology for analyzing and processing language data, and the specific use of the natural language processing technology is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs, such as using a cloud API service to use the natural language processing technology, the Pearson correlation coefficient refers to a statistical index between -1 and 1 for determining whether there is a convergence between the structured data and the pre-structured data, and the value range is [-1, 1], such as obtaining the data cumulative deviation and , the structured data deviation sum , and the pre-structured data deviation sum in the embodiment, set P , the structured data deviation sum refers to the sum of the mean deviation of all data in the structured data, the pre-structured data deviation sum refers to the sum of the mean deviation of all data in the pre-structured data, the mean deviation refers to the difference between a single data and the average value of the data, the data cumulative deviation sum refers to the sum of the product of all structured data deviations and pre-structured data mean deviations, the pre-structured data refers to the structured data in the multi-modal language data within 2 hours before the current time point when the multi-modal language data is received, the structured data and the pre-structured data direction do not converge refers to the case that the semantics between the structured data and the pre-structured data do not converge, the structured data and the pre-structured data direction converge refers to the case that the semantics between the structured data and the pre-structured data converge, the domain database refers to a highly structured special database constructed in the cloud for storing domain knowledge related to the multi-modal language data, and the auxiliary conversion refers to a process of referring to the domain database for analysis and judgment to obtain the literal logic when the natural language processing model encounters key ambiguity, uncertainty and information missing problems in the semantic understanding process.

[0081] Specifically, the logic optimization unit verifies the convergence of the structured data, and adds the domain database auxiliary judgment in the process of converting the multi-modal language data into the literal logic according to the convergence verification result, so as to improve the accuracy of the literal logic.

[0082] Specifically, the logic updating unit analyzes the ambiguity of the unstructured data by the natural language processing technology to obtain an ambiguity analysis result, the ambiguity analysis result includes existence of divergence and non-existence of divergence, and performs semantic validity analysis on the unstructured data according to the ambiguity analysis result, wherein:

[0083] When the ambiguity analysis result is no divergence, no semantic validity analysis is performed on the unstructured data;

[0084] When the ambiguity analysis result is divergence, semantic validity analysis is performed on the unstructured data by natural language processing technology to obtain a validity analysis result, the validity analysis result including valid semantics and invalid semantics;

[0085] The logic updating unit updates the output process of the convergence verification result according to the validity analysis result, wherein:

[0086] When the validity analysis result is invalid semantics, the output process of the convergence verification result is not updated, and the process of converting into literal logic is directly structurally optimized:

[0087] When the validity analysis result is valid semantics, the output process of the convergence verification result is updated: semantic features of the unstructured data are extracted by natural language processing technology to obtain unstructured semantic features, the unstructured semantic features are added to the pre-structured data, and the Pearson correlation coefficient P of the structured data and the pre-structured data is recalculated.

[0088] Specifically, the ambiguity analysis result of no divergence refers to the case of semantic uniformity in the unstructured data, the ambiguity analysis result of divergence refers to the case of non-uniform semantics in the unstructured data, the validity analysis result of invalid semantics refers to the case of irrelevant unstructured data to the generated workflow, the validity analysis result of valid semantics refers to the case of relevant unstructured data to the generated workflow, and the semantic features refer to structured feature data representing the core meaning of the unstructured data extracted from the unstructured data.

[0089] Specifically, the logic updating unit performs ambiguity analysis and semantic validity analysis on the unstructured data to improve the accuracy of the convergence verification result, thereby further improving the accuracy of the literal logic.

[0090] Specifically, the privacy processing module desensitizes the literal logic to obtain desensitized literal logic, and generates a target request fingerprint according to the desensitized literal logic by using a hash function.

[0091] The privacy processing module performs black word standardization processing on the target request fingerprint, constructs a black word standardization processing model, and inputs the target request fingerprint into the black word standardization processing model to obtain a semantic understanding result output by the black word standardization processing model.

[0092] Specifically, the desensitization processing refers to a process of blurring and encrypting sensitive words in the literal logic to protect privacy. The specific implementation of the desensitization processing is not limited in the embodiment, and a person skilled in the art can freely choose according to actual needs, such as constructing a sensitive word rule library by taking the literal logic as an index and taking the sensitive word replacement semantics as the result associated with the literal logic, and performing desensitization processing according to the sensitive word rule library. The hash function refers to a technology of converting the desensitized literal logic into a fixed-length string. The target request fingerprint refers to a fixed-length string that can be recognized by a computer after the desensitized literal logic is converted by the hash function. The slang standardization processing model refers to a BERT-CRF model that takes the target request fingerprint as input data and takes the semantic understanding result as output data. The BERT-CRF model refers to a sequence labeling model architecture for constructing the slang standardization processing model, and the Chinese name is a sequence labeling model based on a BERT pre-training model and a conditional random field. The specific construction method of the slang standardization processing model is not limited in the embodiment, and a person skilled in the art can freely choose according to actual needs, such as training the BERT pre-training model by taking the desensitized literal logic corresponding to the target request fingerprint annotated by a domain expert and its corresponding semantic understanding result as a semantic mapping training set to obtain the slang standardization processing model. The semantic understanding result refers to a logical semantics that conforms to the specific requirements of the business scenario of the workflow generation after the slang standardization processing.

[0093] Specifically, the privacy processing module performs desensitization processing on the literal logic and generates a target request fingerprint to blur and avoid private words in the literal logic, thereby improving data security. The privacy processing module also performs slang standardization processing to make the semantic understanding result conform to the use scenario, thereby improving the accuracy and adaptability of the workflow generation.

[0094] Specifically, the intent understanding module includes:

[0095] The intent generation unit is configured to obtain a semantic confidence according to the semantic understanding result, update the process of the slang standardization processing according to the semantic confidence, and output a first semantic intent.

[0096] The intent processing unit is configured to perform bidirectional verification on the first semantic intent by using a knowledge graph, obtain a bidirectional verification result, and acquire a conflict-free first semantic intent according to the bidirectional verification result. The intent processing unit is connected to the intent generation unit.

[0097] Specifically, the intention generation unit obtains semantic confidence according to the semantic understanding result, updates the process of the slang standardization processing according to the semantic confidence, takes the BERT assignment probability in the slang standardization processing model as the semantic confidence zc, compares the semantic confidence zc with a preset semantic confidence zc0, judges the compliance degree of the semantic confidence according to the comparison result, and updates the process of the slang standardization processing according to the judgment result, wherein:

[0098] When zc≥zc0, the intention generation unit determines that the compliance degree of the semantic confidence is up to standard, does not update the process of the slang standardization processing, and outputs the semantic understanding result as the first semantic intention;

[0099] When zc<zc0, the intention generation unit determines that the compliance degree of the semantic confidence is not up to standard, updates the process of the slang standardization processing, pushes the semantic understanding result to the user through an interactive interface for confirmation, obtains a confirmed semantic understanding result, and outputs the confirmed semantic understanding result as the first semantic intention.

[0100] Specifically, the BERT assignment probability refers to a confidence probability of the slang standardization processing model assigned to each word element in the semantic understanding result, and the preset semantic confidence refers to a preset value for judging the compliance degree of the semantic confidence. The specific numerical value of the preset semantic confidence zc0 is not limited in the embodiment, and can be freely selected by a person skilled in the art according to actual needs. For example, zc0=0.9 is set according to historical experience in the embodiment. The compliance degree of the semantic confidence includes up to standard and not up to standard. The specific pushing manner of the semantic understanding result pushed to the user through the interactive interface is not limited in the embodiment, and can be freely selected by a person skilled in the art according to actual needs. For example, the semantic understanding result is pushed to the user in the form of a question through the interactive interface, and the user selects “yes” or “no” through the interactive interface pushing window.

[0101] Specifically, the intention generation unit judges the compliance degree of the semantic confidence. When the compliance degree of the confidence is not up to standard, the semantic understanding result is timely confirmed to the user, thereby improving the accuracy and adaptability of the workflow.

[0102] Specifically, the intention processing unit performs bidirectional verification on the first semantic intention, constructs a semantic intention knowledge graph, and performs bidirectional verification on the first semantic intention according to the semantic intention knowledge graph to obtain a bidirectional verification result. The bidirectional verification result includes verification passed and verification not passed. According to the bidirectional verification result, a non-conflict first semantic intention is obtained, wherein:

[0103] When the bidirectional verification result is verification pass, the first semantic intent is subjected to semantic conflict resolution to obtain a conflict-free first semantic intent.

[0104] When the bidirectional verification result is verification fail, the first semantic intent is marked as missing and pushed to an administrator.

[0105] Specifically, the semantic intent knowledge graph refers to a knowledge graph taking the first semantic intent as input data and taking a work entity related to the first semantic intent as output. The specific construction method of the semantic intent knowledge graph is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. For example, the first semantic intent-business scenario-cryptic expression-semantic interpretation is taken as an entity node of the knowledge graph, and the relationship type such as synonym relationship-hierarchical relationship-sequential relationship is taken as a relationship type to construct the semantic intent knowledge graph. The bidirectional verification includes forward verification and reverse verification. The forward verification refers to a process of searching for whether there is an entity node corresponding to the first semantic intent in the semantic intent knowledge graph. The reverse verification refers to a process of searching for whether there is a first semantic intent corresponding to the entity node. The bidirectional verification result being verification pass refers to a case that both the forward verification result and the reverse verification result exist in the bidirectional verification result. The bidirectional verification result being verification fail refers to a case that at least one of the forward verification result and the reverse verification result does not exist in the bidirectional verification result. The semantic conflict resolution refers to a process of resolving mutually contradictory semantic interpretations when the system detects the mutually contradictory semantic interpretations. The specific implementation method of the semantic conflict resolution is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. For example, the mutually contradictory semantic interpretations are subjected to weighted voting, and the semantic interpretation with a high weighted voting result is selected. The specific method of pushing to the administrator is not limited in the embodiment, and a person skilled in the art can freely select according to actual needs. For example, an alarm pop-up window is displayed on a management interactive interface.

[0106] Specifically, the intent processing unit verifies the first semantic intent through the knowledge graph bidirectional verification to ensure the execution rationality of the first semantic intent, and resolves the semantic conflict of the first semantic intent according to the first semantic intent to improve the accuracy of the first semantic intent.

[0107] Specifically, the workflow generation module generates a workflow according to the conflict-free first semantic intent through a workflow generation method.

[0108] Specifically, the workflow explanation module explains the target workflow through a workflow explanation method.

[0109] Please refer to Figure 2As shown, it is a flowchart of the workflow generation method of the embodiment, which includes:

[0110] Step A01, obtaining the conflict-free first semantic intention fingerprint according to the conflict-free first semantic intention by a hash function, and writing the conflict-free first semantic intention fingerprint into the blockchain;

[0111] Step A02, generating a workflow Python script according to the conflict-free first semantic intention, and performing cross-platform container execution according to the workflow Python script to obtain a cross-platform execution result, the cross-platform execution result including execution success and execution failure;

[0112] Step A03, performing execution compensation on the process of cross-platform container execution according to the cross-platform execution result, wherein:

[0113] When the cross-platform execution result is execution success, no execution compensation is performed on the process of cross-platform container execution;

[0114] When the cross-platform execution result is execution failure, performing execution compensation on the process of cross-platform container execution: all business impacts and resource changes generated by the workflow Python script of cross-platform container execution in the execution process.

[0115] Specifically, the conflict-free first semantic intention fingerprint refers to converting the conflict-free first semantic intention into a fixed-length string recognizable by a computer through a hash function, the writing into the blockchain refers to the process of permanently and irreversibly storing the first semantic intention fingerprint record into the distributed ledger technology, the workflow Python script refers to the Python program code dynamically generated according to the conflict-free first semantic intention for executing the workflow, the workflow refers to the process of executing specific work business, the cross-platform container execution refers to the technical solution of packaging the workflow Python script into a container environment, running consistently on different operating systems and hardware platforms, and checking whether a reasonable workflow can be generated, the cross-platform execution result for execution success refers to the case where the workflow Python script successfully generates a workflow after cross-platform container execution, the cross-platform execution result for execution failure refers to the case where the workflow Python script fails to successfully generate a workflow after cross-platform container execution, the business impact refers to the changes caused by the workflow Python script to business data, state and process, and the resource change refers to the entities and computing elements created, modified and occupied by the workflow Python script in cross-platform execution, such as application programs pulled and deployed from the code repository.

[0116] Specifically, the workflow generation module permanently saves while being traceable by writing the conflict-free first semantic intention fingerprint into the blockchain, while the workflow generation module generates the workflow Python script and performs cross-platform container execution to verify the feasibility of cross-platform execution of the workflow Python script, thereby improving the platform adaptability of the workflow Python script.

[0117] Referring to Figure 3 As shown in the flowchart of the work explanation method, the work explanation method comprises:

[0118] Step B01, generating a Chinese briefing diagram according to the cross-platform execution result through natural language processing technology;

[0119] Step B02, pushing the Chinese briefing diagram to the user end and the operation end, acquiring the user like amount Ds through the user end, and performing differential privacy test on the Chinese briefing diagram through the operation end to obtain a privacy score Ys;

[0120] Step B03, comparing the user like amount Ds with a preset user like amount Ds0, judging the compliance degree of the user like amount according to the comparison result, and outputting a first privacy explanation decision according to the judgment result, wherein:

[0121] When Ds is greater than or equal to Ds0, it is determined that the compliance degree of the user like amount is up to standard, and the first privacy explanation decision is output: new slang is mined from enterprise chat records and meeting minutes, and the new slang is added to the training library of the slang standardization processing model to retrain the slang standardization processing model;

[0122] When Ds is less than Ds0, it is determined that the compliance degree of the user like amount is not up to standard, and the first privacy explanation decision is output: opinion feedback is pushed to the user through an interactive interface, and the opinion feedback result is sent to the operation and maintenance personnel;

[0123] Step B04, comparing the privacy score Ys with a preset privacy score Ys0, setting the preset privacy score Ys0=0.85, judging the compliance degree of the privacy score according to the comparison result, and outputting a second privacy explanation decision according to the judgment result, wherein:

[0124] When Ys is greater than or equal to Ys0, it is determined that the compliance degree of the privacy score is up to standard, and the second privacy explanation decision is output: the conflict-free first semantic intention is marked as an old intention, the availability of the historical old intention is automatically checked, and the user is reminded to update the system according to the availability of the historical old intention;

[0125] When Ys < Ys0, it is determined that the compliance degree of the privacy score is unqualified, and the second privacy explanation decision is output: the operation end supplements the entity node in the semantic intention knowledge graph to obtain a supplemented semantic intention knowledge graph, and re-performs bidirectional verification on the first semantic intention according to the supplemented semantic intention knowledge graph.

[0126] Specifically, the Chinese briefing figure refers to an easy-to-understand text description converted from a complex technical cross-platform execution result through natural language processing technology, the differential privacy refers to an existing privacy protection technology that provides strict mathematical proof by adding controllable noise, the privacy score refers to a numerical value measuring the privacy degree of the workflow Python script, and the epsilon value in the differential privacy is taken as the privacy score in the embodiment, the preset user like amount refers to a preset value for judging the compliance degree of the user like amount, which can be set according to platform historical experience, the preset privacy score refers to a preset value for judging the compliance degree of the privacy score, and Ys0=0.85 is set according to the standard threshold of the differential privacy in the embodiment, the opinion feedback refers to an opinion table for pushing and recording the opinions of users, and the compliance degree of the privacy score includes qualified and unqualified.

[0127] Specifically, the workflow explanation module judges the compliance degree of the user like amount and the compliance degree of the privacy score, so as to ensure the workflow matching and privacy of the workflow Python script generated by the workflow explanation module, and outputs the first privacy explanation decision and the second privacy explanation decision, so as to further improve the privacy protection of the workflow generation.

[0128] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without departing from the principles of the present application, and the technical solutions after the changes or replacements will fall within the protection scope of the present application.

Claims

1. A natural language based cross-platform workflow intelligent generation and interpretation system, characterized in that, The system comprises: a language receiving module configured to receive multi-modal language data, the multi-modal language data comprising structured data and unstructured data; a language conversion module configured to convert the multi-modal language data into literal logic, and to structure-optimize the process of converting into literal logic, and to update the results of the process of structure optimization and the process of converting into literal logic; a privacy processing module configured to desensitize the literal logic to obtain desensitized literal logic, and to generate a target request fingerprint according to the desensitized literal logic, and to perform underworld standardization processing on the target request fingerprint to obtain a semantic understanding result; an intent understanding module configured to output a first semantic intent according to the semantic understanding result, and to obtain a non-conflict first semantic intent according to the first semantic intent; a workflow generation module configured to obtain a target semantic fingerprint according to the non-conflict first semantic intent, and to generate a target workflow according to the target semantic fingerprint; a workflow interpretation module configured to interpret and optimize the workflow according to the target workflow.

2. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 1, wherein, The language conversion module comprises: a logic conversion unit configured to convert the multi-modal language data into literal logic; a logic optimization unit configured to perform convergence verification on the structured data to obtain a convergence verification result, and to structure-optimize the process of converting into literal logic according to the convergence verification result; a logic updating unit configured to perform ambiguity analysis on the unstructured data to obtain an ambiguity analysis result, and to perform semantic validity analysis on the unstructured data according to the ambiguity analysis result to obtain a validity analysis result, and to update the output process of the convergence verification result according to the validity analysis result.

3. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 2, wherein, The logic conversion unit converts the multi-modal language data into literal logic through natural language processing technology; The logic optimization unit performs convergence verification on the structured data, calculates the Pearson correlation coefficient P of the structured data and the previous structured data through Python, and -1 When -1 When 0 When the logic optimization unit structure-optimizes the process of converting into literal logic, the process of converting the multi-modal language data into literal logic is assisted through a domain database.

4. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 3, wherein, The logic updating unit performs ambiguity analysis on the unstructured data through natural language processing technology to obtain an ambiguity analysis result, the ambiguity analysis result comprising existence of divergence and non-existence of divergence, and performs semantic validity analysis on the unstructured data according to the ambiguity analysis result, wherein: When the ambiguity analysis result is non-existence of divergence, the unstructured data is not subjected to semantic validity analysis; When the ambiguity analysis result is a divergence, semantic validity analysis is performed on the unstructured data by natural language processing technology to obtain a validity analysis result, the validity analysis result including valid semantics and invalid semantics; The logic updating unit updates the output process of the convergence verification result according to the validity analysis result, wherein: When the validity analysis result is invalid semantics, the output process of the convergence verification result is not updated, and the process of converting into literal logic is directly structurally optimized: When the validity analysis result is valid semantics, the output process of the convergence verification result is updated: semantic features of the unstructured data are extracted by natural language processing technology to obtain unstructured semantic features, the unstructured semantic features are added to the pre-structured data, and the Pearson correlation coefficient P of the structured data and the pre-structured data is recalculated.

5. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 4, wherein, The privacy processing module desensitizes the literal logic to obtain desensitized literal logic, and generates a target request fingerprint according to the desensitized literal logic by a hash function; The privacy processing module performs slang standardization processing on the target request fingerprint, constructs a slang standardization processing model, and inputs the target request fingerprint into the slang standardization processing model to obtain a semantic understanding result output by the slang standardization processing model.

6. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 5, wherein, The intent understanding module includes: An intent generation unit configured to obtain semantic confidence according to the semantic understanding result, update the process of slang standardization processing according to the semantic confidence, and output a first semantic intent; An intent processing unit configured to perform bidirectional verification of the knowledge graph on the first semantic intent to obtain a bidirectional verification result, and obtain a conflict-free first semantic intent according to the bidirectional verification result.

7. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 6, wherein, The intent generation unit obtains semantic confidence according to the semantic understanding result, updates the process of slang standardization processing according to the semantic confidence, takes the BERT assignment probability in the slang standardization processing model as the semantic confidence zc, compares the semantic confidence zc with a preset semantic confidence zc0, judges the compliance degree of the semantic confidence according to the comparison result, and updates the process of slang standardization processing according to the judgment result, wherein: When zc≥zc0, the intent generation unit determines that the compliance degree of the semantic confidence is up to standard, does not update the process of slang standardization processing, and outputs the semantic understanding result as the first semantic intent; When zc<zc0, the intent generation unit determines that the compliance degree of the semantic confidence is not up to standard, updates the process of slang standardization processing: the semantic understanding result is pushed to the user through an interactive interface for confirmation to obtain a confirmed semantic understanding result, and the confirmed semantic understanding result is output as the first semantic intent.

8. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 7, wherein, The intention processing unit performs knowledge graph bidirectional verification on the first semantic intention, constructs a semantic intention knowledge graph, and performs bidirectional verification on the first semantic intention according to the semantic intention knowledge graph to obtain a bidirectional verification result, the bidirectional verification result including verification pass and verification fail, and the non-conflict first semantic intention is obtained according to the bidirectional verification result, wherein: When the bidirectional verification result is verification pass, the semantic conflict of the first semantic intention is resolved to obtain the non-conflict first semantic intention; When the bidirectional verification result is verification fail, the first semantic intention is marked as missing and pushed to an administrator.

9. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 8, wherein, The workflow generation module generates a workflow according to the non-conflict first semantic intention by a workflow generation method, the workflow generation method comprising: Step A01, obtaining a non-conflict first semantic intention fingerprint according to the non-conflict first semantic intention by a hash function, and writing the non-conflict first semantic intention fingerprint into a blockchain; Step A02, generating a workflow Python script according to the non-conflict first semantic intention, and performing cross-platform container execution according to the workflow Python script to obtain a cross-platform execution result, the cross-platform execution result including execution success and execution failure; Step A03, performing execution compensation on the process of cross-platform container execution according to the cross-platform execution result, wherein: When the cross-platform execution result is execution success, no execution compensation is performed on the process of cross-platform container execution; When the cross-platform execution result is execution failure, execution compensation is performed on the process of cross-platform container execution: all business impacts and resource changes generated by the workflow Python script executed by the cross-platform container during the execution process.

10. The natural language based cross-platform workflow intelligent generation and interpretation system according to claim 9, wherein, The workflow interpretation module interprets the target workflow by a workflow interpretation method, the workflow interpretation method comprising: Step B01, generating a Chinese briefing graph according to the cross-platform execution result by natural language processing technology; Step B02, pushing the Chinese briefing graph to a user end and an operation end, obtaining a user like amount Ds through the user end, and performing differential privacy testing according to the Chinese briefing graph through the operation end to obtain a privacy score Ys; Step B03, comparing the user like amount Ds with a preset user like amount Ds0, judging the compliance degree of the user like amount according to the comparison result, and outputting a first privacy interpretation decision according to the judgment result, wherein: When Ds≥Ds0, it is determined that the compliance degree of the user like amount is up to standard, and the first privacy interpretation decision is output: new slang is mined from enterprise chat records and meeting minutes, and the new slang is added to the training library of the slang standardization processing model to retrain the slang standardization processing model; When Ds<Ds0, it is determined that the compliance degree of the user like amount is not up to standard, and the first privacy interpretation decision is output: opinion feedback is pushed to the user through an interactive interface, and the opinion feedback result is sent to an operation and maintenance personnel; Step B04, comparing the privacy score Ys with a preset privacy score Ys0, setting the preset privacy score Ys0=0.85, judging the compliance degree of the privacy score according to the comparison result, and outputting the second privacy explanation decision according to the judgment result, wherein: When Ys≥Ys0, it is determined that the compliance degree of the privacy score is up to standard, and the second privacy explanation decision is output: the conflict-free first semantic intention is marked as an old intention, the availability of the historical old intention is automatically checked, and the user is reminded to update the system according to the availability of the historical old intention; When Ys<Ys0, it is determined that the compliance degree of the privacy score is not up to standard, and the second privacy explanation decision is output: the operating end supplements the entity node in the semantic intention knowledge graph to obtain a supplemented semantic intention knowledge graph, and re-verifies the first semantic intention according to the supplemented semantic intention knowledge graph.

Images