Policy document pushing method based on label calculation, medium and equipment

Abstract

The invention discloses a policy document pushing method based on label calculation, a medium and equipment, and the method comprises the steps: obtaining the basic information of each enterprise, inputting the basic information of each enterprise into a trained first neural network model, and obtaining a first label corresponding to each enterprise; obtaining a plurality of policy files, slicing each policy file to obtain a plurality of text slices, and inputting all the text slices into a trained second neural network model to obtain a second label corresponding to each text slice; according to the matching degree of the first label and the second label, establishing a mapping relationship between the text slice and each enterprise; when the first condition is met, the text slices are pushed to the corresponding enterprises according to the mapping relation. In this way, the text slices of the policy document can be pushed to the corresponding enterprises more accurately, and the suitability and accuracy of the recommended content are improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a policy document push method, medium and device based on label calculation. Background Art

[0002] Traditional government and enterprise recommendation systems require the configuration of many matching rules, which are complex in rule design and require high customization. Each new rule needs to be redeveloped, which results in poor flexibility and high matching difficulty. In addition, the policy documents pushed to enterprises are often in the form of complete articles, which are usually lengthy, making it impossible for enterprises to accurately obtain the policy content that is most closely related to themselves, affecting search query efficiency. Summary of the invention

[0003] In view of the above problems, this application provides a technical solution for policy document push based on label calculation to solve the problem that the existing policy document push method cannot accurately match some policy content required by the enterprise to the enterprise, affecting the timeliness and accuracy of the enterprise's acquisition of policy information.

[0004] To achieve the above objectives, in a first aspect, the present application provides a policy file push method based on tag calculation, the method comprising:

[0005] Obtaining basic information of each enterprise, inputting the basic information of the enterprise into the trained first neural network model, and obtaining a first label corresponding to each enterprise;

[0006] Acquire multiple policy documents, slice each of the policy documents to obtain a plurality of text slices, input all the text slices into a trained second neural network model, and obtain a second label corresponding to each text slice;

[0007] According to the matching degree between the first label and the second label, a mapping relationship between the text slice and each enterprise is established;

[0008] When the first condition is met, the text slice is pushed to the corresponding enterprise according to the mapping relationship. The first condition includes any one of the following: policy document update, the time from the last push reaching a preset interval, and receiving a policy query instruction.

[0009] Furthermore, when pushing the text slice to the corresponding enterprise, the method further includes:

[0010] The link of the policy document to which the current text slice belongs is pushed to the corresponding enterprise, and when a click instruction for the link of the policy document is received from the enterprise, the content of the policy document is displayed.

[0011] Furthermore, the content of the policy document to which the display belongs includes:

[0012] The content corresponding to the text slice in the policy file is highlighted.

[0013] Furthermore, the method comprises:

[0014] Detect in real time whether the basic information of each enterprise is updated. If so, re-input the updated basic information of the enterprise into the trained first neural network model to obtain the updated first label corresponding to each enterprise;

[0015] Alternatively, it is detected in real time whether the policy document is updated. If so, the updated policy document is re-sliced ​​to obtain new text slices, and the new text slices are input into the trained second neural network model to obtain a second label corresponding to the new text slices.

[0016] Furthermore, the method further comprises:

[0017] According to the matching degree between the first tag and the second tag, a push weight value of a corresponding text slice is set for each enterprise in the mapping relationship;

[0018] After each time the text slice is pushed to the corresponding enterprise, feedback information from the enterprise regarding the text slice pushed this time is received, the push weight value corresponding to the text slice pushed this time is adjusted according to the feedback information, and the adjusted push weight value is used to push the text slice to the enterprise during the next push.

[0019] Furthermore, each of the policy documents is sliced ​​to obtain a number of text slices including:

[0020] Determine the initial vector length according to the total length of the policy document and the segmentation accuracy;

[0021] S61: Slice the policy document according to the initial vector length to obtain initial text slices;

[0022] S62: Acquire a plurality of post-paragraph sentences after the current initial text slice T, and analyze the correlation between the plurality of post-paragraph sentences and the current initial text slice T and the correlation between the plurality of post-paragraph sentences;

[0023] S63: determining a first segmentation adjustment strategy according to the strength of the association between the plurality of post-paragraph sentences and the current initial text slice T and the strength of the association between the plurality of post-paragraph sentences, performing segmentation adjustment on the current initial text slice T according to the first segmentation adjustment strategy, and updating the segmentation pointer position;

[0024] S64: Segment the next text slice starting from the updated segmentation pointer position;

[0025] Repeat steps S61-S64 until the current policy document is completely segmented to obtain a plurality of text slices.

[0026] Further, the plurality of post-paragraph statements include a first post-paragraph statement H1 and a second post-paragraph statement H2, and the correlation between T and H1 is R1, the correlation between H1 and H2 is R2, and the correlation between T and H2 is R3;

[0027] Step S63 includes:

[0028] If it is determined that R1 is not a strong association relationship, the current initial text slice T is kept unchanged, and the segmentation pointer position is updated to the end position of the current initial text slice T;

[0029] If it is determined that R1 is a strong association relationship but R2 is a non-strong association relationship, H1 is included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H1;

[0030] If it is determined that R1 and R2 are both strongly associated but R3 is not strongly associated, then both H1 and H2 are included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H1;

[0031] If it is determined that R1, R2, and R3 are all strongly associated, H1 and H2 are both included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H2.

[0032] Furthermore, if it is determined that R1, R2, and R3 are all strongly associated, before H1 and H2 are included in the current initial text slice T, the following steps are also included:

[0033] S81: obtaining a plurality of end-of-segment sentences located at the end of the current initial text slice T, and analyzing the correlation between the plurality of end-of-segment sentences and H1 and the correlation between the plurality of end-of-segment sentences;

[0034] S82: determining a second segmentation adjustment strategy according to the correlation between the end-of-segment sentences and H1 and the correlation between the end-of-segment sentences;

[0035] S83: Determine whether to include H1 and H2 into the current initial text slice T according to the second segmentation adjustment strategy;

[0036] The plurality of end-of-segment statements include a first end-of-segment statement Q1 and a second end-of-segment statement Q2, wherein the correlation between Q1 and H1 is R1', the correlation between Q1 and Q2 is R5', and the correlation between Q2 and H1 is R6';

[0037] Steps S82-S83 include:

[0038] If it is determined that R1' is not a strong association, the current initial text slice T is kept unchanged, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of the current initial text slice T;

[0039] If it is determined that R1' is a strong correlation but R2' is a non-strong correlation, Q1 is removed from the current initial text slice T, Q1, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of H2;

[0040] If it is determined that R1' and R2' are both strongly associated but R3' is not strongly associated, the current initial text slice T is kept unchanged, H1, H2, Q1 and Q2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of H2, or Q1 and Q2 are removed from the current initial text slice T, Q1, Q2, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of H2;

[0041] If it is determined that R1', R2', and R3' are all strongly associated, H1 and H2 are both included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H2.

[0042] In a second aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the policy file push method based on label calculation as described in the first aspect of the present application.

[0043] In a third aspect, the present application provides an electronic device having a computer program stored thereon, including a processor and a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by the processor, the policy file push method based on label calculation as described in the first aspect of the present application is implemented.

[0044] Different from the prior art, the above technical solution involves a policy document push method, medium and device based on label calculation, the method comprising: obtaining basic information of each enterprise, inputting the basic information of the enterprise into a trained first neural network model, and obtaining a first label corresponding to each enterprise; obtaining multiple policy documents, slicing each policy document, obtaining several text slices, inputting all text slices into a trained second neural network model, and obtaining a second label corresponding to each text slice; establishing a mapping relationship between the text slice and each enterprise according to the matching degree between the first label and the second label; when the first condition is met, pushing the text slice to the corresponding enterprise according to the mapping relationship. In the above manner, the text slices of the policy document can be pushed to the corresponding enterprise more accurately, improving the adaptability and accuracy of the recommended content.

[0045] The above-mentioned records related to the invention content are only an overview of the technical solution of the present application. In order to enable ordinary technicians in the field to more clearly understand the technical solution of the present application, and then implement it according to the text of the specification and the contents recorded in the drawings, and to make the above-mentioned purpose and other purposes, features and advantages of the present application easier to understand, the following is an explanation in combination with the specific implementation mode and drawings of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings are only used to illustrate the principles, implementation methods, applications, characteristics and effects of the specific embodiments of the present invention and other related contents, and shall not be considered as limiting the present application.

[0047] In the drawings of the specification:

[0048] Figure 1 This is a flowchart of the policy file push method based on tag calculation described in the first exemplary embodiment of the present application;

[0049] Figure 2 This is a flow chart of the policy file push method based on tag calculation described in the second exemplary embodiment of the present application;

[0050] Figure 3 This is a flow chart of the policy file push method based on tag calculation described in the third exemplary embodiment of the present application;

[0051] Figure 4 This is a flow chart of the policy file push method based on tag calculation described in the fourth exemplary embodiment of the present application;

[0052] Figure 5 A schematic diagram of text slicing involved in the first exemplary embodiment of the present application;

[0053] Figure 6 A schematic diagram of an electronic device according to an exemplary embodiment of the present application;

[0054] The reference numerals in the above drawings are described as follows:

[0055] 10. Electronic equipment;

[0056] 101. Processor;

[0057] 102. Storage medium. DETAILED DESCRIPTION

[0058] In order to explain in detail the possible application scenarios, technical principles, specific schemes that can be implemented, and the purposes and effects that can be achieved, the following is a detailed description of the specific embodiments listed in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0059] Reference to "embodiment" herein means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present application. The term "embodiment" appearing in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or association with other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the various technical features mentioned in the embodiments can be combined in any way to form a corresponding implementable technical solution.

[0060] Unless otherwise defined, the technical terms used in this document have the same meanings as those generally understood by those skilled in the art to which this application belongs; the use of relevant terms in this document is only for describing specific embodiments and is not intended to limit this application.

[0061] In the description of this application, the term "and / or" is an expression used to describe the logical relationship between objects, indicating that three relationships may exist, for example, A and / or B, which means: A exists, B exists, and A and B exist at the same time. In addition, the character " / " in this article generally indicates that the objects before and after are in an "or" logical relationship.

[0062] In the present application, terms such as “first” and “second” are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship of quantity, priority or sequence between these entities or operations.

[0063] Without further limitations, in this application, the words "include", "comprises", "has" or other similar open-ended expressions used in sentences are intended to cover non-exclusive inclusion. These expressions do not exclude the presence of additional elements in the process, method or product including the elements, so that the process, method or product including a series of elements may include not only those limited elements, but also other elements not explicitly listed, or also include elements inherent to such process, method or product.

[0064] Similar to the understanding in the Examination Guidelines, in this application, expressions such as "greater than", "less than", "exceed" and the like are understood to exclude the number itself; expressions such as "above", "below", "within" and the like are understood to include the number itself. In addition, in the description of the embodiments of this application, "multiple" means more than two (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups", "multiple times", etc., unless otherwise clearly and specifically limited.

[0065] In the description of the embodiments of the present application, space-related expressions used, such as "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicate the orientation or position relationship based on the orientation or position relationship shown in the specific embodiments or drawings, and are only for the convenience of describing the specific embodiments of the present application or facilitating the reader's understanding, and do not indicate or imply that the referred device or component must have a specific position, a specific orientation, or be constructed or operated in a specific orientation, and therefore cannot be understood as a limitation on the embodiments of the present application.

[0066] Unless otherwise expressly specified or limited, in the description of the embodiments of the present application, the terms such as "install", "connect", "connect", "fix", "set", etc. used should be understood in a broad sense. For example, the "connection" can be a fixed connection, a detachable connection, or an integrated setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium; it can be the internal connection of two elements or the interaction relationship between two elements. For technicians in the technical field to which the present application belongs, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0067] In the first aspect, Figure 1 As shown, the present application provides a policy file push method based on label calculation, the method comprising:

[0068] First, proceed to step S101 to obtain basic information of each enterprise, input the basic information of the enterprise into the trained first neural network model, and obtain a first label corresponding to each enterprise;

[0069] Then proceed to step S102 to obtain multiple policy documents, slice each of the policy documents to obtain a number of text slices, input all the text slices into the trained second neural network model, and obtain a second label corresponding to each text slice;

[0070] Then, step S103 is performed to establish a mapping relationship between the text slice and each enterprise according to the matching degree between the first label and the second label;

[0071] Then, the process proceeds to step S104, and when the first condition is met, the text slice is pushed to the corresponding enterprise according to the mapping relationship.

[0072] In step S101, the basic information of the enterprise includes the business information of the enterprise user (such as the enterprise name, business scope, registered capital, paid-in capital), etc., the industry field to which it belongs, the browsing time of policy documents, the type of policy public account followed, the number of searches, downloads, and sharing of policy documents, financial data (operating income, profit, assets and liabilities) and operating risks (such as guarantee information, equity pledge), and contact information (such as email address, mobile phone number).

[0073] During training, the input of the first neural network model can be the basic data of the sample enterprise, and the output is the first label corresponding to each enterprise. The first label is used to characterize the policy portrait of the enterprise, including the policy document types that the enterprise is concerned about and the policy document types that may be used in the enterprise operation process.

[0074] In step S102, the input of the second neural network model can be the sample text slices corresponding to the sample policy document, and the output is the second label corresponding to each sample text slice. The second label can be obtained by summarizing the semantic analysis of the sample text slice, that is, the most relevant keywords representing the text slice paragraph.

[0075] The first neural network model and the second neural network model can be a convolutional neural network (CNN) or a recurrent neural network (RNN) and their variants (such as LSTM, GRU).

[0076] In step S103, based on the degree of match between the first tag and the second tag, it can be determined whether the text slice in a policy document needs to be pushed to the corresponding enterprise in a timely manner. The higher the degree of match between the first tag and the second tag, the greater the probability that the current text slice will be pushed to the corresponding enterprise.

[0077] In step S104, the first condition includes any one of: policy file update, time from last push reaching a preset interval, and policy query instruction received.

[0078] The above scheme first divides the relatively lengthy policy document into text slices, and then performs semantic analysis on the text slices, assigns a corresponding second label to each text slice, and based on the matching degree between the second label and the first label (obtained by analyzing the basic information of the enterprise), when the first condition is met, the text slice of the policy document is pushed to the corresponding enterprise, which can improve the adaptability and accuracy of the recommended policy content.

[0079] In some embodiments, when pushing the text slice to the corresponding enterprise, the method also includes: pushing a link to the policy file to which the current text slice belongs to the corresponding enterprise, and displaying the content of the policy file when receiving a click instruction from the enterprise on the link to the policy file.

[0080] In this way, when the enterprise receives the corresponding text slice, if it wants to further view the full text of the policy document corresponding to the text slice, it can directly view it by clicking on the pushed link without searching on the web page, which effectively improves the speed at which the enterprise obtains policy information.

[0081] In some embodiments, the display of the content of the policy file includes: highlighting the content corresponding to the text slice in the policy file. The highlighting refers to adding different marking indicators to make the position of the pushed text slice in the entire policy file more noticeable. The marking indicator can be a mark of the text slice filled with different colors, or a frame of the text slice with a specific graphic.

[0082] By highlighting the content corresponding to the text slice, enterprise users can more intuitively and clearly know the location of the text slice when viewing the complete policy document.

[0083] In some embodiments, the method comprises:

[0084] Detect in real time whether the basic information of each enterprise is updated. If so, re-input the updated basic information of the enterprise into the trained first neural network model to obtain the updated first label corresponding to each enterprise;

[0085] Alternatively, it is detected in real time whether the policy document is updated. If so, the updated policy document is re-sliced ​​to obtain new text slices, and the new text slices are input into the trained second neural network model to obtain a second label corresponding to the new text slices.

[0086] In short, the first label and the second label are not static. As the basic information of the enterprise changes, the first label can also be re-evaluated to make the first label more accurate. Preferably, the first label is determined by selecting the basic information of the enterprise within a preset time interval from the current timestamp and inputting it into the first neural network model. As for the second label, a dynamic maintenance table of the second label corresponding to a policy document can be organized. Every time a policy document is updated, if it is a revision of a previously existing policy document, only the text slices corresponding to the policy documents updated in the most recent few times will be maintained to ensure the timeliness of the second label. If the policy document updated this time is a newly issued policy document, a corresponding second label set will be created, and the second label corresponding to the text slice of the policy document will be maintained.

[0087] In some embodiments, Figure 2 As shown, the method also includes:

[0088] First, enter step S201, according to the matching degree between the first tag and the second tag, set the push weight value of the corresponding text slice for each enterprise in the mapping relationship;

[0089] Then enter step S202, after each time the text slice is pushed to the corresponding enterprise, receive the enterprise's feedback information on the text slice pushed this time, adjust the push weight value corresponding to the text slice pushed this time according to the feedback information, and use the adjusted push weight value to push the text slice to the enterprise in the next push.

[0090] In this embodiment, the feedback information refers to the feedback result of the degree of help of the text slice pushed by the enterprise user this time. If the feedback information tends to be helpful, the corresponding text slice will be actively pushed to the enterprise next time the policy document of this type is updated. On the contrary, if the feedback information tends to be not very helpful, the push frequency of the corresponding text slice will be actively reduced next time the policy document of this type is updated. In this way, the recommendation of text slices can be made more accurate and meet the needs of enterprise development.

[0091] In some embodiments, Figure 3 As shown, each of the policy documents is sliced ​​to obtain a number of text slices including:

[0092] S60: determining the initial vector length according to the total length of the policy document and the segmentation accuracy;

[0093] S61: Slice the policy document according to the initial vector length to obtain initial text slices;

[0094] S62: Acquire a plurality of post-paragraph sentences after the current initial text slice T, and analyze the correlation between the plurality of post-paragraph sentences and the current initial text slice T and the correlation between the plurality of post-paragraph sentences;

[0095] S63: determining a first segmentation adjustment strategy according to the strength of the association between the plurality of post-paragraph sentences and the current initial text slice T and the strength of the association between the plurality of post-paragraph sentences, performing segmentation adjustment on the current initial text slice T according to the first segmentation adjustment strategy, and updating the segmentation pointer position;

[0096] S64: Segment the next text slice starting from the updated segmentation pointer position;

[0097] Repeat steps S61-S64 until the current policy document is completely segmented to obtain a plurality of text slices.

[0098] In step S60, the segmentation accuracy may include two accuracy parameters, namely, a coherence parameter and a search matching accuracy parameter. The coherence parameter has a value range of 0-1, 1 represents the most coherent, 0 represents the least coherent, and the initial value may be set to 0.5 by default. The search matching accuracy parameter has a value range of 0-1, 1 represents the most accurate match, 0 represents the most fuzzy match, and the initial value may be set to 0.5 by default. The initial vector length may have a value range of 100-300, the minimum segment length is 100, and the maximum segment length is 300.

[0099] In step S62, the strength of the correlation between the post-paragraph sentences and the current initial text slice T, and the strength of the correlation between the post-paragraph sentences can be calculated by a semantic similarity algorithm based on text vectorization.

[0100] Through the above method, the classification of the post-paragraph sentences can be made more accurate, and the content relevance in each adjusted text slice T can be stronger, which is convenient for generating a more accurate second label, thereby improving the adaptability of the text slice push of the policy document.

[0101] In some embodiments, the plurality of post-paragraph statements include a first post-paragraph statement H1 and a second post-paragraph statement H2, and the correlation between T and H1 is R1, the correlation between H1 and H2 is R2, and the correlation between T and H2 is R3;

[0102] Step S63 includes:

[0103] S631: If it is determined that R1 is not a strong association relationship, the current initial text slice T is kept unchanged, and the segmentation pointer position is updated to the end position of the current initial text slice T;

[0104] S632: If it is determined that R1 is a strong association relationship but R2 is a non-strong association relationship, H1 is included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H1;

[0105] S633: If it is determined that R1 and R2 are both strongly associated but R3 is not strongly associated, then both H1 and H2 are included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H1;

[0106] S634: If it is determined that R1, R2, and R3 are all strongly associated, H1 and H2 are both included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H2.

[0107] In short, the situation of step S631 shows that the sentence after the current initial text slice T (i.e., H1) is not closely related to the current initial text slice T, so the current initial text slice T can be kept unchanged, and the next slice starts from text slice T. The situation of step S632 shows that the sentence after the current initial text slice T (i.e., H1) is strongly related to the current initial text slice T, but the next sentence H2 is not closely related to H1. In order to make the semantic relevance in the text slice stronger, H1 needs to be included in the text slice T, and the next slice starts from the end of H1. The situation of step 633 shows that there is a strong correlation between the current initial text slice T and the next sentence H1, H1 and H2 (the sentence after H1), but H2 is weakly related to the text slice T. At this time, H1 and H2 will be included in the text slice T, but the next segmentation pointer starts from the end of H1. At this time, H2 is equivalent to a transition sentence between two adjacent slices, which exists in both the current initial text slice and the next text slice. The situation in step 634 indicates that there is a strong correlation between the current initial text slice T and H1 and H2. At this time, the text slice T and H1 and H2 can be regarded as a whole. H1 and H2 are both included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H2, that is, the next text slice is sliced ​​starting from the end position of H2.

[0108] In some embodiments, Figure 4 As shown, if it is determined that R1, R2, and R3 are all strongly associated, before H1 and H2 are included in the current initial text slice T, the following is also included:

[0109] S81: obtaining a plurality of end-of-segment sentences located at the end of the current initial text slice T, and analyzing the correlation between the plurality of end-of-segment sentences and H1 and the correlation between the plurality of end-of-segment sentences;

[0110] S82: determining a second segmentation adjustment strategy according to the correlation between the end-of-segment sentences and H1 and the correlation between the end-of-segment sentences;

[0111] S83: Determine whether to include H1 and H2 into the current initial text slice T according to the second segmentation adjustment strategy;

[0112] like Figure 5 As shown, the several end-of-segment statements include the first end-of-segment statement Q1 and the second end-of-segment statement Q2. The correlation between Q1 and H1 is R1', the correlation between Q1 and Q2 is R5', and the correlation between Q2 and H1 is R6';

[0113] Steps S82-S83 include:

[0114] If it is determined that R1' is not a strong association, the current initial text slice T is kept unchanged, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of the current initial text slice T;

[0115] If it is determined that R1' is a strong correlation but R2' is a non-strong correlation, Q1 is removed from the current initial text slice T, Q1, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of H2;

[0116] If it is determined that R1' and R2' are both strongly associated but R3' is not strongly associated, the current initial text slice T is kept unchanged, H1, H2, Q1 and Q2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of H2, or Q1 and Q2 are removed from the current initial text slice T, Q1, Q2, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of H2;

[0117] If it is determined that R1', R2', and R3' are all strongly associated, H1 and H2 are both included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H2.

[0118] Through the above method, the length of text slices can be made more balanced, which is convenient for subsequent analysis. At the same time, the correlation between sentences in the same text slice can be made closer, thereby improving the accuracy of the subsequent generation of the second label corresponding to each text slice.

[0119] In a second aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the policy file push method based on label calculation as described in the first aspect of the present invention is implemented.

[0120] The computer-readable storage medium may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories.

[0121] The non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a ferromagnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD ROM); the magnetic surface memory can be a disk memory or a tape memory.

[0122] The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), direct memory bus random access memory (DRRAM). The computer-readable storage medium described in the embodiments of the present invention is intended to include these and any other suitable types of memory.

[0123] like Figure 6 As shown, in a third aspect, the present invention provides an electronic device 10, comprising a processor 101 and a storage medium 102, wherein a computer program is stored on the storage medium, and when the computer program is executed by the processor, the policy file push method based on label calculation as described in the first aspect of the present invention is implemented.

[0124] In some embodiments, the processor can be implemented by software, hardware, firmware or a combination thereof, and can use circuits, single or multiple application specific integrated circuits (Application Specific Integrated Circuit, ASIC), digital signal processors (Digital Signal Processor, DSP), digital signal processing devices (Digital Signal Processing Device, DSPD), programmable logic devices (Programmable Logic Device, PLD), field programmable gate arrays (Field Programmable Gate Array, FPGA), central processing units (Central Processing Unit, CPU), controllers, microcontrollers, microprocessors and at least one thereof, so that the processor can execute some or all of the steps or any combination of the steps in the policy file push method based on label calculation described in the various embodiments of the present application.

[0125] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of this application, this does not limit the scope of patent protection of this application. All technical solutions generated by replacing or modifying equivalent structures or equivalent processes based on the essential concept of this application using the contents recorded in the specification and drawings of this application, as well as directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are included in the scope of patent protection of this application.

Claims

1. A policy document push method based on tag calculation, characterized in that: The method comprises: Obtaining basic information of each enterprise, inputting the basic information of the enterprise into the trained first neural network model, and obtaining a first label corresponding to each enterprise; Acquire multiple policy documents, slice each of the policy documents to obtain a plurality of text slices, input all of the text slices into a trained second neural network model, and obtain a second label corresponding to each text slice; Establishing a mapping relationship between the text slice and each enterprise according to the matching degree between the first label and the second label; When the first condition is met, the text slice is pushed to the corresponding enterprise according to the mapping relationship. The first condition includes any one of the following: policy document update, time from the last push reaching a preset interval, and receiving a policy query instruction.

2. A policy document push method based on tag calculation according to claim 1, characterized in that: When pushing the text slice to the corresponding enterprise, the method further includes: The link of the policy document to which the current text slice belongs is pushed to the corresponding enterprise, and when a click instruction on the link of the policy document is received from the enterprise, the content of the policy document is displayed.

3. A policy document push method based on tag calculation as claimed in claim 2, characterized in that: The content of the policy document showing the ownership includes: The content corresponding to the text slice in the policy file is highlighted.

4. The policy document push method based on tag calculation according to claim 1, characterized in that: The method comprises: Real-time detection of whether the basic information of each enterprise is updated. If so, the updated basic information of the enterprise is re-inputted into the trained first neural network model to obtain the updated first label corresponding to each enterprise; Alternatively, it is detected in real time whether the policy document is updated. If so, the updated policy document is re-sliced ​​to obtain a new text slice, and the new text slice is input into the trained second neural network model to obtain a second label corresponding to the new text slice.

5. The policy document push method based on tag calculation according to claim 1, characterized in that: The method further comprises: According to the matching degree between the first tag and the second tag, a push weight value of the corresponding text slice is set for each enterprise in the mapping relationship; After each time the text slice is pushed to the corresponding enterprise, feedback information from the enterprise regarding the text slice pushed this time is received, the push weight value corresponding to the text slice pushed this time is adjusted according to the feedback information, and the adjusted push weight value is used to push the text slice to the enterprise during the next push.

6. The policy document push method based on tag calculation according to claim 1, characterized in that: Each of the policy documents is sliced ​​to obtain several text slices including: S60: Determine the initial vector length according to the total length of the policy document and the segmentation accuracy; S61: Slice the policy document according to the length of the initial vector to obtain initial text slices; S62: Acquire a plurality of post-paragraph statements following the current initial text slice T, and analyze the relevance between the plurality of post-paragraph statements and the current initial text slice T, and the relevance between the plurality of post-paragraph statements; S63: Determine a first segmentation adjustment strategy based on the strength of the correlation between the plurality of post-segment statements and the current initial text slice T, and the strength of the correlation between the plurality of post-segment statements, perform segmentation adjustment on the current initial text slice T according to the first segmentation adjustment strategy, and update the segmentation pointer position; S64: Segment the next text slice starting from the updated segmentation pointer position; Repeat steps S61-S64 until all the current policy documents are segmented to obtain a plurality of text slices.

7. The policy file push method based on tag calculation according to claim 6, characterized in that: The plurality of post-paragraph statements include a first post-paragraph statement H1 and a second post-paragraph statement H2, wherein the correlation between T and H1 is R1, the correlation between H1 and H2 is R2, and the correlation between T and H2 is R3; Step S63 includes: If it is determined that R1 is not a strong association relationship, the current initial text slice T is kept unchanged, and the segmentation pointer position is updated to the end position of the current initial text slice T; If it is determined that R1 is a strong association relationship but R2 is a non-strong association relationship, H1 is included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H1; If it is determined that R1 and R2 are both strongly associated but R3 is not strongly associated, then both H1 and H2 are included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H1; If it is determined that R1, R2, and R3 are all strongly associated, then H1 and H2 are both included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H2.

8. The policy file push method based on tag calculation according to claim 7, characterized in that: If it is determined that R1, R2, and R3 are all strongly associated, before H1 and H2 are included in the current initial text slice T, the following steps are also included: S81: Acquire several end-of-segment sentences at the end of the current initial text slice T, and analyze the relevance between the several end-of-segment sentences and H1 and the relevance between the several end-of-segment sentences; S82: determining a second segmentation adjustment strategy based on the correlation between the end-of-segment sentences and H1 and the correlation between the end-of-segment sentences; S83: Determine whether to include H1 and H2 in the current initial text slice T according to the second segmentation adjustment strategy; The plurality of end-of-segment statements include the first end-of-segment statement Q1 and the second end-of-segment statement Q2. The correlation between Q1 and H1 is R1', the correlation between Q1 and Q2 is R5', and the correlation between Q2 and H1 is R6'. Steps S82-S83 include: If it is determined that R1' is not a strong association, the current initial text slice T is kept unchanged, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of the current initial text slice T; If it is determined that R1' is a strong association relationship but R2' is a non-strong association relationship, Q1 is removed from the current initial text slice T, Q1, H1 and H2 are merged and sliced ​​independently, and the segmentation pointer position is updated to the end position of H2; If it is determined that R1' and R2' are both strongly associated but R3' is not strongly associated, then keep the current initial text slice T unchanged, merge H1, H2, Q1 and Q2 into an independent slice, and update the segmentation pointer position to the end position of H2, or remove Q1 and Q2 from the current initial text slice T, merge Q1, Q2, H1 and H2 into an independent slice, and update the segmentation pointer position to the end position of H2; If it is determined that R1', R2', and R3' are all strongly associated, H1 and H2 are both included in the current initial text slice T, and the segmentation pointer position is updated to the end position of H2.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the policy file pushing method based on label calculation as described in any one of claims 1 to 8 is implemented.

10. An electronic device having a computer program stored thereon, characterized in that: The method comprises a processor and a storage medium, wherein a computer program is stored on the storage medium, and when the computer program is executed by the processor, the policy file push method based on label calculation as described in any one of claims 1 to 8 is implemented.

Images