Method and device for generating suggestion information based on enterprise information semantic understanding

Abstract

Embodiments of the present disclosure disclose a suggestion information generation method and device based on enterprise information semantic understanding. A specific implementation of the method comprises: performing dynamic file segmentation on an enterprise information file submitted by a target enterprise terminal according to a summary item feature set to obtain a set of information file blocks; performing cross semantic feature extraction on each information file block in the set of information file blocks to generate file block semantic features corresponding to the information file blocks; performing data pre-filling on a summary information table according to the set of information file blocks and the file block semantic features corresponding to the information file blocks to generate a first summary information table and a second summary information table; generating first suggestion information according to the first summary information table; and generating second suggestion information according to the second summary information table. The implementation not only guarantees semantic integrity, but also guarantees the accuracy of subsequent suggestion generation.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the fields of computer technology, artificial intelligence, and machine learning, and specifically to a method and apparatus for generating suggestion information based on semantic understanding of enterprise information. Background Technology

[0002] "Assisted suggestion generation" refers to a technology that combines artificial intelligence to automate data analysis and suggestion generation. Taking the enterprise application field as an example, the traditional method usually requires converting relevant policy requirement data into judgment rules in advance, and then processing the application content through conditional judgment.

[0003] However, the above methods often have the following problems: (1.1) When there is a lot of data required by relevant policies or when the data required by relevant policies changes, the conditional judgment method requires the maintenance of a large number of judgment rules, which is extremely costly. (1.2) The conditional judgment method can only give the judgment result of whether it is satisfied or not, and cannot give corresponding processing suggestions based on the enterprise information documents provided by the enterprise.

[0004] While there are some cases where AI-based suggestions have been used to address the aforementioned issues, they have also created new problems: (2.1) Due to the limitation of lexical units, it is often impossible to parse large amounts of enterprise information files at once. In particular, the sequential segmentation method may sever the contextual relationship between data, thereby affecting feature extraction and data understanding, and thus affecting the accuracy of suggestion generation. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a method and apparatus for generating suggestion information based on semantic understanding of enterprise information, in order to solve the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a method for generating suggestion information based on semantic understanding of enterprise information. The method includes: dynamically segmenting an enterprise information file submitted by a target enterprise terminal according to a set of summary item features to obtain a set of information file blocks, wherein the summary item features correspond to summary items in a summary information table, and the information file blocks correspond to at least one summary item in the summary information table; performing cross-semantic feature extraction on each information file block in the set of information file blocks to generate file block semantic features corresponding to the information file block, wherein the file block semantic features consist of a main semantic feature and a secondary semantic feature group; and performing suggestion information generation based on the information file blocks... The semantic features of the file blocks corresponding to the collection and information file blocks are used to pre-fill the above-mentioned summary information table to generate a first summary information table and a second summary information table. The first summary information table represents the summary items that have been filled in the above summary information table, and the second summary information table represents the summary items that have not been filled in the above summary information table. Based on the first summary information table, first suggestion information is generated, which represents verification suggestions for the summary items in the first summary information table. Based on the second summary information table, second suggestion information is generated, which represents supplementary suggestions for the summary items in the second summary information table.

[0008] Secondly, some embodiments of this disclosure provide a suggestion information generation apparatus based on enterprise information semantic understanding. The apparatus includes: a dynamic file segmentation unit configured to dynamically segment enterprise information files submitted by a target enterprise terminal according to a set of summary item features to obtain a set of information file blocks, wherein the summary item features correspond to summary items in a summary information table, and the information file blocks correspond to at least one summary item in the summary information table; a cross-semantic feature extraction unit configured to extract cross-semantic features from each information file block in the set of information file blocks to generate file block semantic features corresponding to the information file blocks, wherein the file block semantic features consist of a main semantic feature and a secondary semantic feature group; and a data pre-filling unit configured to... Based on the aforementioned set of information file blocks and the semantic features of the corresponding file blocks, the aforementioned summary information table is pre-filled with data to generate a first summary information table and a second summary information table. The first summary information table represents the filled-in summary items in the aforementioned summary information table, and the second summary information table represents the unfilled summary items in the aforementioned summary information table. A first generation unit is configured to generate first suggestion information based on the aforementioned first summary information table, wherein the first suggestion information represents verification suggestions for the summary items in the aforementioned first summary information table. A second generation unit is configured to generate second suggestion information based on the aforementioned second summary information table, wherein the second suggestion information represents supplementary suggestions for the summary items in the aforementioned second summary information table.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0011] The above embodiments of this disclosure have the following beneficial effects: The suggestion information generation method based on enterprise information semantic understanding in some embodiments of this disclosure achieves automated and highly accurate suggestion information generation. Specifically, based on the summary item feature set, the enterprise information file submitted by the target enterprise terminal is dynamically segmented to obtain an information file block set. The summary item features correspond to the summary items in the summary information table, and the information file block corresponds to at least one summary item in the summary information table. In practice, due to word limitations, it is often impossible to parse large amounts of enterprise information files at once. Conventional methods mainly use sequential segmentation, but this segmentation method may sever the contextual association between data, thus affecting feature extraction and data understanding. Therefore, this disclosure performs dynamic file segmentation by combining the summary item features corresponding to the summary items in the summary information table. This ensures the semantic integrity within the information file block and enables the association between the information file block and the summary items in the summary information table during the segmentation stage. Next, cross-semantic feature extraction is performed on each information file block in the above information file block set to generate file block semantic features corresponding to the above information file block. The file block semantic features consist of a main semantic feature and a secondary semantic feature group. In practice, although dynamic segmentation improves the semantic integrity of each information file block, since enterprise information files are a holistic description of the target enterprise terminal, certain semantic relationships still exist between the information file blocks. Therefore, this disclosure uses cross-semantic feature extraction to capture the semantic relationships between features of information file blocks, thereby improving the expressive power of features. Next, based on the aforementioned set of information file blocks and the semantic features of the corresponding file blocks, the aforementioned summary information table is pre-filled to generate a first summary information table and a second summary information table. The first summary information table represents the filled summary items in the first summary information table, and the second summary information table represents the unfilled summary items in the second summary information table. In practice, since the association between information file blocks and summary items in the summary information table is achieved during the dynamic file segmentation stage, by combining the extracted semantic features of the corresponding file blocks, automatic matching and filling of summary items in the semantic understanding dimension can be achieved. Compared with conditional judgment-based processing methods, this significantly improves matching accuracy. Furthermore, based on the aforementioned first summary information table, first suggestion information is generated, wherein the first suggestion information represents a verification suggestion for the summary items in the first summary information table. Finally, based on the aforementioned second summary information table, second suggestion information is generated, wherein the second suggestion information represents supplementary suggestions for the summary items in the second summary information table. By combining the filled-in summary items and the unfilled summary items, corresponding verification suggestions and supplementary suggestions are automatically generated.In summary, the semantic understanding of enterprise information eliminates the need to set up and maintain a large number of judgment rules, thus avoiding the extremely high maintenance costs associated with conditional judgment-based processing methods. Furthermore, the dynamic file segmentation method, which combines summary item features, not only ensures semantic integrity but also guarantees the accuracy of subsequent suggestion generation. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 These are flowcharts of some embodiments of the suggestion information generation method based on semantic understanding of enterprise information according to this disclosure; Figure 2 This is a schematic diagram illustrating the generation process of summary item features; Figure 3 This is a schematic diagram illustrating the calculation process of vector values ​​in the matching vector corresponding to the descriptive sentence features; Figure 4 This is a schematic diagram of the structure of some embodiments of the suggestion information generation apparatus based on semantic understanding of enterprise information according to the present disclosure; Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] Before performing any of the operations related to the collection, storage, and use of corporate information documents disclosed in this disclosure, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, informing personal information subjects, and obtaining prior authorization and consent from personal information subjects.

[0020] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] refer to Figure 1 The diagram illustrates a flow 100 of some embodiments of a suggestion information generation method based on enterprise information semantic understanding according to the present disclosure. This suggestion information generation method based on enterprise information semantic understanding includes the following steps: Step 101: Based on the feature set of the summary items, dynamically segment the enterprise information file submitted by the target enterprise terminal to obtain a set of information file blocks.

[0022] In some embodiments, the execution entity (e.g., a computing device) of the suggestion information generation method based on enterprise information semantic understanding can dynamically segment the enterprise information file submitted by the target enterprise terminal according to the summary item feature set to obtain an information file block set.

[0023] The summary item features correspond to the summary items in the summary information table. The information file block corresponds to at least one summary item in the summary information table. The target enterprise terminal is a mobile terminal bound to the target enterprise terminal account. The target enterprise is the enterprise to be certified. For example, the target enterprise could be a high-tech enterprise or a specialized, refined, and innovative enterprise. The summary information table is a data table in summary form set up according to the certification requirements. The summary item can represent the certification item corresponding to the certification requirements. The summary item features can represent the semantic features corresponding to the summary item. The information file block is a file block obtained by dynamically segmenting the enterprise information file. The enterprise information file can be enterprise-related information provided by the target enterprise through the target enterprise terminal for the corresponding enterprise certification requirements. For example, taking the specialized, refined, and innovative enterprise certification project as an example, the enterprise information file can include, but is not limited to: enterprise name, unified social credit code, enterprise registered address, enterprise mailing address, postal code, business registration date, registered capital, enterprise size type, legal representative and contact information, enterprise industry, enterprise sub-sector type, enterprise type, introduction of main products, introduction of economic benefits and operating conditions, introduction of enterprise honors, introduction of enterprise qualifications, supporting materials, etc.

[0024] In practice, different summary information tables are often set up for different certification requirements. Therefore, semantic features can be extracted in advance from the summary items in the summary information tables corresponding to different certification requirements. For example, text encoding can be performed using the Word2Vec model to generate summary item features corresponding to the summary items. Specifically, the set of summary item features corresponding to the certification item selected by the target enterprise terminal can be extracted.

[0025] As an example, see Figure 2 The diagram illustrates the process of generating summary item features. Taking identified items A1, A2, and A3 as examples, the summary information tables corresponding to identified items A1, A2, and A3 contain different numbers and types of summary items. First, the executing entity can pre-encode the summary items in the summary information tables corresponding to identified items A1, A2, and A3 using the Word2Vec model, respectively, to obtain summary item feature sets D1, D2, and D3. Next, identified item A1 is associated with summary item feature set D1 (e.g., through the item number corresponding to identified item A1), identified item A2 is associated with summary item feature set D2, and identified item A3 is associated with summary item feature set D3, and these associations are stored in the summary item feature database. The summary item feature database can be a vector database specifically designed for storing summary item features.

[0026] In practice, the process of generating a set of information file blocks may include the following steps: Step S1: Segment the paragraphs within the aforementioned enterprise information document to obtain a set of information paragraphs. Specifically, paragraph segmentation can be performed using the heading structure within the enterprise information document as the dividing point.

[0027] Step S2: For each information segment in the above information segment set, when the number of words corresponding to the information segment is greater than the preset word threshold, the above information segment is equally divided at the sentence level to obtain the segmented information segment; when the number of words corresponding to the information segment is less than or equal to the preset word threshold, the information segment is taken as the segmented information segment.

[0028] Among them, the preset token threshold represents the maximum number of tokens processed in a single transaction.

[0029] For example, assuming a preset word threshold of M and a word count of M+10 for an information segment, the conventional approach is to divide the information segment into two parts: one with M words and the other with 10 words. However, the latter, with its smaller word count, makes it difficult to extract sufficient semantic features. Furthermore, segmentation within a sentence can compromise the semantic integrity of that sentence. Therefore, the equivalence segmentation method used in this disclosure ensures that the segmented information segments contain sufficient semantic features, and the sentence-level granularity, being the segmentation granularity, also guarantees the semantic integrity of the sentence.

[0030] Step S3: Determine the semantic features of each segmented information segment.

[0031] Specifically, word embedding features can be extracted from the segmented information paragraphs using a model such as text-embedding-3-small to obtain the paragraph semantic features corresponding to the segmented information paragraphs.

[0032] Step S4: For each summary item feature, determine the K segmented information paragraphs that match the summary item feature by calculating similarity (calculating the summary item feature and paragraph semantic features).

[0033] Step S5: When the total number of words in the K segmented information paragraphs corresponding to the semantic features of the summary item is less than or equal to the preset word threshold, the K segmented information paragraphs are concatenated to form an information file block; when the total number of words in the K segmented information paragraphs corresponding to the semantic features of the summary item is greater than the preset word threshold, the concatenated result of the K segmented information paragraphs is divided into multiple information file blocks using sentence granularity as the segmentation granularity and the preset word threshold as the segmentation restriction.

[0034] In particular, there may be multiple summary items corresponding to the same information file block. Therefore, it is necessary to merge and associate the summary items and information file blocks to avoid duplicate storage of information file blocks.

[0035] It should be noted that the aforementioned computing devices can be either hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device. When the computing device is software, it can be installed on the hardware devices listed above. It can be implemented as, for example, multiple software programs or software modules used to provide distributed services, or as a single software program or software module. No specific limitations are made here.

[0036] In some optional implementations of some embodiments, the above-described dynamic file segmentation of the enterprise information file submitted by the target enterprise terminal based on the digest item feature set yields a set of information file blocks, including: Step S1: Based on the enterprise information file, perform the following processing steps: Step S11: Based on the preset word threshold, the enterprise information file is segmented into file blocks to obtain the initial information file blocks.

[0037] The preset token threshold represents the maximum number of tokens processed in a single operation. The initial information file block is located at the beginning of the enterprise information file and contains no more than or equal to the preset token threshold. The initial file block consists of a sequence of descriptive sentences. Each descriptive sentence is a sentence-level statement contained within the initial file.

[0038] In practice, the aforementioned execution entity can use sentence granularity as the segmentation granularity to sequentially segment the enterprise information document from beginning to end. When the total number of words corresponding to the M descriptive sentences obtained by segmentation is greater than or equal to the preset word element threshold and the total number of words corresponding to the first M-1 descriptive sentences is less than or equal to the preset word element threshold, the first M-1 descriptive sentences are used as the descriptive sentence sequence included in the initial information file block.

[0039] Step S12: Determine the descriptive features corresponding to each descriptive sentence in the descriptive sentence sequence to obtain the descriptive sentence feature sequence.

[0040] The descriptive sentence features include: semantic features and semantic keyword features. Semantic keyword features represent the keyword features corresponding to the semantic keywords contained within the descriptive sentence. The semantic features of the descriptive sentence are the overall semantic expression of the descriptive sentence. Semantic keywords are the keywords within the descriptive sentence. Semantic keyword features are the semantic expressions corresponding to the keywords.

[0041] In practice, firstly, the semantic features of the descriptive sentence can be extracted using the text-embedding-3-small model. Then, by combining these semantic features with part-of-speech tagging, at least one keyword corresponding to the descriptive sentence is identified. Next, local features corresponding to the keyword are extracted from the semantic features of the descriptive sentence to serve as the semantic keyword features.

[0042] Step S13: For each descriptive sentence feature in the descriptive sentence feature sequence, perform feature matching between the descriptive sentence feature and the above summary item feature set to generate a matching vector corresponding to the descriptive sentence feature.

[0043] The matching vector represents the degree of matching between the description sentence and the feature set of the summary items. Assuming that the feature set of the summary items contains N summary item features, the vector dimension of the matching vector can be 1×N.

[0044] In practice, for each summary item feature in the summary item feature set, the aforementioned execution entity can calculate the feature similarity (cosine similarity) between the semantic features of the description sentence included in the description sentence feature and the summary item feature, as well as the feature similarity (cosine similarity) between the semantic keyword features in the semantic keyword feature group included in the description sentence feature and the summary item feature. The obtained multiple feature similarities are then weighted and averaged as the vector value in the matching vector corresponding to the description sentence feature that corresponds to the aforementioned summary item feature.

[0045] As an example, see Figure 3 The diagram illustrates the calculation process of vector values ​​in the matching vector corresponding to the descriptive sentence features. Taking descriptive sentence feature P1 as an example, descriptive sentence feature P1 includes: descriptive sentence semantic feature P11 and semantic keyword feature group P12. Semantic keyword feature group P12 includes: semantic keyword feature P121 and semantic keyword feature P122. The summary item feature set includes: summary item feature D11, summary item feature D12, ..., summary item feature D1n. Therefore, the vector dimension of the matching vector can be 1×N, and the matching vector can include: the vector value corresponding to summary item feature D11, the vector value corresponding to summary item feature D12, ..., the vector value corresponding to summary item feature D1n. Taking the calculation of the vector value corresponding to summary item feature D11 as an example, firstly, the aforementioned execution entity calculates the feature similarity between summary item feature D11 and descriptive sentence semantic feature P11, the feature similarity between summary item feature D11 and semantic keyword feature P121, and the feature similarity between summary item feature D11 and semantic keyword feature P122. Then, the aforementioned execution entity averages the similarity of the three features, using this average as the vector value in the matching vector corresponding to the descriptive sentence feature P1 that corresponds to the summary item feature D11. Wherein, Figure 3 The “Sim” in the text indicates a similarity calculation between features (e.g., cosine similarity calculation).

[0046] Step S14: Based on the obtained matching vector sequence, filter the description sentence sequence to obtain the filtered description sentence sequence, which serves as the information file block.

[0047] In practice, firstly, since each matching vector represents the feature similarity between a description sentence and each feature of the summary item feature set, the top K (e.g., K is set to 2 by default) summary items with the highest matching degree can be determined statistically based on the vector values ​​in the matching vector sequence. For example, assuming the description sentence sequence contains M-1 description sentences, the vector dimension of the matching vector sequence can be (M-1)×N. The vertical dimension of the matching vector sequence represents the vector values ​​between the same summary item feature and each description sentence. Therefore, the overall similarity between the information file block (description sentence sequence) and the summary item feature can be determined by average weighting. Thus, the overall similarity between the information file block and each summary item feature can be obtained as (1×N), allowing the selection of the Top-K summary items with the highest matching degree (maximum overall similarity). Then, for each matching vector in the matching vector sequence, if the vector values ​​corresponding to these K highest-matching summary items are not among the Top 3 vector values ​​in the matching vector sequence, the description sentence corresponding to that matching vector is removed.

[0048] Step S15: In response to the fact that the enterprise information file after removing the filtered description sequence is empty, the above processing steps are terminated.

[0049] Step S2: In response to the fact that the enterprise information file after removing the filtered description sequence is not empty, the enterprise information file after removing the filtered description sequence is used as the enterprise information file, and the above processing steps are executed again.

[0050] In practice, this method not only meets the limit on the maximum number of tokens processed in a single operation, but also solves the problem of semantic integrity that is difficult to guarantee in traditional sequential segmentation by using the features of the summary item as a reference.

[0051] Step 102: Perform cross-semantic feature extraction on each information file block in the information file block set to generate the file block semantic features corresponding to the information file block.

[0052] In some embodiments, the execution entity described above may perform cross-semantic feature extraction on each information file block in the information file block set to generate file block semantic features corresponding to the information file block.

[0053] The semantic features of a file block consist of primary semantic features and secondary semantic features. The primary semantic features represent the overall semantic features corresponding to the aforementioned information file block. The secondary semantic features in the secondary semantic feature group represent the semantic features of two adjacent information file blocks and the cross-semantic features between the aforementioned information file blocks.

[0054] In practice, since the above-mentioned information file block segmentation is not performed sequentially, but rather based on the summary item features to ensure semantic integrity, there are still certain semantic relationships between the information file blocks in the resulting information file block set, especially between adjacent information file blocks. In order to capture these semantic relationships and improve the robustness of feature expression, a cross-semantic feature extraction method is used to further capture semantic features.

[0055] Specifically, firstly, since each information file block is equivalent to a corresponding filtered description sentence sequence, and each description sentence has corresponding semantic features, the semantic features of the filtered description sentences included in the information file block can be sequentially concatenated as the initial file block features. Then, the initial file block features are compressed to become the main semantic features of the information file block. Next, the two main semantic features of two adjacent information file blocks and the main semantic feature of the current information file block are concatenated (front and back) and then compressed (e.g., using a text-embedding-3-small model for secondary encoding compression) to become the secondary semantic features in the secondary semantic feature group.

[0056] As an example, consider information file blocks 1, 2, and 3. First, the main semantic features corresponding to information file block 1 and information file block 2 are concatenated and then compressed to obtain secondary semantic features. Next, the main semantic features corresponding to information file block 2 and information file block 3 are concatenated and then compressed to obtain secondary semantic features. This yields two secondary semantic features (secondary semantic feature groups) corresponding to each information file block.

[0057] In some optional implementations of certain embodiments, the above-mentioned cross-semantic feature extraction of each information file block in the above-mentioned information file block set to generate file block semantic features corresponding to the above-mentioned information file block includes: Step S1: Determine the semantic feature sequence of the first description sentence, the semantic feature sequence of the second description sentence, and the semantic feature sequence of the third description sentence.

[0058] Wherein, the first descriptive sentence semantic feature sequence is the descriptive sentence semantic feature sequence corresponding to the information file block, the second descriptive sentence semantic feature sequence is the descriptive sentence semantic feature sequence corresponding to the preceding information file block, and the third descriptive sentence semantic feature sequence is the descriptive sentence semantic feature sequence corresponding to the following information file block. The preceding information file block is located before and adjacent to the aforementioned information file block, and the following information file block is located after and adjacent to the aforementioned information file block.

[0059] In practice, since each information file block is equivalent to a corresponding filtered sequence of descriptive sentences, and each descriptive sentence has its own semantic features, the semantic features of the multiple descriptive sentences corresponding to the aforementioned information file blocks can first be used as the first sequence of descriptive sentence semantic features. Next, the semantic features of the multiple descriptive sentences corresponding to the information file blocks preceding and adjacent to the aforementioned information file blocks are used as the second sequence of descriptive sentence semantic features. Finally, the semantic features of the multiple descriptive sentences corresponding to the information file blocks following and adjacent to the aforementioned information file blocks are used as the third sequence of descriptive sentence semantic features.

[0060] Step S2: Perform global semantic feature extraction on the semantic feature sequence of the first description sentence, the semantic feature sequence of the second description sentence, and the semantic feature sequence of the third description sentence respectively to obtain the semantic features of the first file block, the semantic features of the second file block, and the semantic features of the third file block.

[0061] In practice, the semantic features of the first, second, and third file blocks are generated in the same way. Therefore, taking the extraction process of the semantic features of the first file block as an example, firstly, a fully connected layer is used to map the semantic features of the first descriptive sentence in the first descriptive sentence semantic feature sequence, obtaining the first mapped feature sequence. This achieves feature dimension compression and feature shaping. The feature dimension of the first mapped features can be 768 dimensions. Secondly, based on the sentence position of the corresponding descriptive sentence in the corresponding information file block, sine and cosine positional encoding is used to obtain the positional encoding corresponding to the first mapped features. Next, each first mapped feature in the first mapped feature sequence is concatenated with its corresponding positional encoding to obtain the first concatenated feature sequence. Finally, a standard Transformer encoder is used to encode the first concatenated feature sequence as a whole to obtain the semantic features of the first file block. This method achieves feature compression and semantic association based on the semantic features of the descriptive sentence, thereby capturing the feature expressions of multiple adjacent information file blocks.

[0062] Step S3: Determine the semantic features of the first file block as the main semantic features included in the semantic features of the file block corresponding to the information file block.

[0063] Step S4: Extract associated semantic features from the semantic features of the first file block and the second file block, as well as from the semantic features of the first file block and the third file block, to obtain the semantic feature group included in the semantic features of the file block corresponding to the information file block.

[0064] In practice, firstly, a cross-attention encoder can be used to perform cross-attention encoding on the semantic features of the first and second file blocks to obtain a semantic feature. Then, the cross-attention encoder can be used to extract associated semantic features from the semantic features of the first and third file blocks respectively. The cross-attention encoder is an encoder that employs a cross-attention mechanism. Compared to traditional attention mechanism encoders, in this disclosure, the semantic features of the first file block are used as a query sequence, and the semantic features of the second file block are used as a key and value sequence to achieve cross-attention encoding of the semantic features of the first and second file blocks. Simultaneously, the semantic features of the first file block are used as a query sequence, and the semantic features of the third file block are used as a key and value sequence to achieve cross-attention encoding of the semantic features of the first and third file blocks. By combining the cross-attention encoder, the associated semantic capture between vector information file blocks is achieved.

[0065] Step 103: Based on the information file block set and the semantic features of the corresponding file blocks, pre-fill the summary information table to generate the first summary information table and the second summary information table.

[0066] In some embodiments, the aforementioned execution entity may pre-fill data into the summary information table based on the information file block set and the semantic features of the file blocks corresponding to the information file blocks, so as to generate a first summary information table and a second summary information table.

[0067] The first summary information table represents the summary items that have been filled in the summary information table, and the second summary information table represents the summary items that have not been filled in the summary information table.

[0068] In practice, firstly, since each information file block corresponds to at least one summary item, a cue vector can be constructed based on this at least one summary item. Then, the cue vector and the semantic features of the corresponding file block are concatenated, and this is used as input to output the summary item fill content corresponding to at least one summary item through a sequence labeling model. The sequence labeling model can employ an LSTM (Long Short Term Memory) + CRF (Conditional Random Fields) labeling model. Next, the summary item fill content is filled into the corresponding fill position in the summary information table. Finally, the summary information table is filtered by the filled summary items to obtain the first summary information table, and the unfilled summary items in the summary information table are filtered to obtain the second summary information table.

[0069] In some optional implementations of certain embodiments, the execution entity pre-fills the summary information table with data based on the information file block set and the semantic features of the file blocks corresponding to the information file blocks, to generate a first summary information table and a second summary information table, including: Step S1: For each information file block in the information file block set, perform the following pre-filling steps: Step S11: Generate a first matching result based on the main semantic features corresponding to the above information file block and the summary item features of at least one summary item corresponding to the above information file block.

[0070] In practice, since an information file block corresponds to at least one summary item, the summary item features of that summary item can be treated as a cue vector. This vector is then concatenated with the main semantic features of the information file block and used as model input. Sequence labeling is then used to locate the summary fill content corresponding to the summary item. Alternatively, a Bi-LSTM (Bidirectional Long Short-Term Memory) + CRF (Conditional Random Fields) labeling model can be used for sequence labeling. The candidate summary item fill content corresponding to at least one summary item of the information file block is then obtained as the first matching result.

[0071] Step S12: Generate a second matching result based on the semantic feature group corresponding to the above information file block and the summary item features of at least one summary item corresponding to the above information file block.

[0072] In practice, since an information file block corresponds to at least one summary item, the summary item features of at least one summary item corresponding to the information file block can be regarded as a cue vector. This cue vector is then concatenated with the semantic feature group corresponding to the information file block as model input, and the summary filling content corresponding to the summary item is located using sequence labeling. Alternatively, a Bi-LSTM (Bidirectional Long Short-Term Memory) + CRF (Conditional Random Fields) labeling model can be used for sequence labeling. The candidate summary item filling content corresponding to at least one summary item corresponding to the information file block is obtained as the second matching result. Specifically, compared with the process of determining the first matching result in step S12, both core methods use sequence labeling to extract the summary item content. However, since the main semantic features are only the semantic expression corresponding to the information file block, they have higher feature convergence. Semantic features, on the other hand, fuse the semantic expressions corresponding to two adjacent information file blocks based on the main semantic features. Although this weakens the semantic expression corresponding to a single information file block, it contains the feature correlation between adjacent information file blocks. Therefore, this disclosure locates the summary item content from these two dimensions to improve the robustness of the extraction. Furthermore, the reason for not using the main semantic features and secondary semantic features as a whole input is that the secondary semantic features contain a certain expression of the main semantic features. Subsequent feature fusion is equivalent to a second superposition and fusion on the basis of the main semantic features, which will weaken the expression of the main semantic features and thus affect the recognition accuracy.

[0073] Step S13: Perform correlation verification on the first matching result and the second matching result to generate a third matching result.

[0074] The third matching result includes: summary item name, summary item fill content, and fill confidence.

[0075] In practice, because the first and second matching results depend on different features, the populated content for the same summary item may differ. This mainly manifests in the following situations: Case (1): The content filled in the summary items corresponding to the same summary item in the first matching result and the second matching result is the same. Therefore, the content filled in the summary item can be directly used as the content filled in the summary item corresponding to the summary item in the third matching result. The confidence scores corresponding to the content filled in the summary item in the first matching result and the second matching result are weighted on average to be used as the filling confidence score.

[0076] Case (2): The fill content of the summary item corresponding to the same summary item in the first matching result and the second matching result is different. In this case, compare the confidence of the fill content of the two different summary items for the same summary item in the first matching result and the second matching result, and take the fill content of the summary item with the highest confidence as the fill content of the summary item, and take the highest confidence as the fill confidence.

[0077] Furthermore, for scenario (2), since the first and second matching results depend on different features, confidence weights can be set for the first and second matching results respectively. For example, the confidence weight of the first matching result can be 0.6, and the confidence weight of the second matching result can be 0.4. When the content filled in the two summary items for the same summary item in the first and second matching results is inconsistent, the confidence weight of the first matching result and the confidence level of the content filled in the summary item are multiplied together to obtain a new confidence level. The confidence weight of the second matching result and the confidence level of the content filled in the summary item are multiplied together to obtain a new confidence level. Then, by comparing the confidence levels, the content filled in the summary item corresponding to the highest confidence level is taken as the content filled in the summary item of that summary item, and the highest confidence level is taken as the filling confidence level.

[0078] By combining the processing methods of cases (1) and (2), the content to be filled in the summary items corresponding to multiple summary items can be determined.

[0079] Step S14: Based on the third matching result above, populate at least one summary item in the summary information table corresponding to the information file block above with summary items.

[0080] In practice, the summary items can be automatically filled in by using multiple preset regular expressions based on the third matching result, so as to fill the corresponding summary item filling position with the summary item content.

[0081] Step S2: Generate the first summary information table based on the summary items already filled in the summary information table.

[0082] In practice, the already filled-in summary items in the summary information table can be extracted and used as the first summary information table.

[0083] Step S3: Generate a second summary information table based on the unfilled summary items in the summary information table. In practice, unfilled summary items in the summary information table can be extracted and used as a second summary information table.

[0084] Step 104: Generate first suggestion information based on the first summary information table.

[0085] In some embodiments, the aforementioned executing entity may generate first suggestion information based on a first summary information table.

[0086] The first suggestion information represents the verification suggestion for the summary item in the first summary information table mentioned above.

[0087] In practice, since the summary items in the first summary information table have been populated with content, and each summary item has a population confidence level, multiple confidence intervals can be set to divide the summary items in the first summary information table into multiple summary item groups, and different verification suggestions can be set according to the confidence interval corresponding to each summary item group.

[0088] In some optional implementations of certain embodiments, the execution entity generates first suggestion information based on the first summary information table, including: Step S1: Filter the first summary information table to obtain the first summary item group, the second summary item group, the third summary item group, and the fourth summary item group.

[0089] The first summary item is a summary item whose corresponding fill confidence score is greater than the first confidence threshold; the second summary item is a summary item whose corresponding fill confidence score is less than or equal to the first confidence threshold but greater than the second confidence threshold; the third summary item is a summary item whose corresponding fill confidence score is less than or equal to the second confidence threshold; and the fourth summary item is a summary item whose fill content has an anomaly with its corresponding summary item. Specifically, the first confidence threshold can be 0.95, the second confidence threshold can be 0.75, and the third confidence threshold can be 0.55. Furthermore, each summary item has multiple pre-set anomaly triggering rules, such as rules for judging abnormal value ranges. Therefore, based on the multiple anomaly triggering rules corresponding to a summary item, it can be determined whether the summary item is the fourth summary item.

[0090] Step S2: Generate batch confirmation suggestions based on the first summary item group mentioned above.

[0091] In practice, because the first summary item has a high confidence level, batch confirmation suggestions can be generated to guide enterprises in performing batch verification. Specifically, the batch confirmation suggestions can include: the position of the first summary item in the first summary information table and a suggested prompt. The suggested prompt could be, "Please perform batch confirmation on the above multiple summary fill-in contents."

[0092] Step S3: Generate batch verification suggestions based on the second summary item group mentioned above.

[0093] The aforementioned batch verification suggestion includes: the location of the first file block and verification auxiliary information. The location of the first file block indicates the file position of the information file block corresponding to the second summary item in the aforementioned enterprise information file. The verification auxiliary information may include: the location of the second summary item within the first summary information table and a suggested guiding text. The suggested guiding text may be, "Please conduct batch verification of the above multiple summary fill-in contents."

[0094] In practice, although the confidence level of the second summary item is relatively high, there is still a possibility that the target enterprise's terminal data may be incorrect, so it is necessary to guide batch review.

[0095] Step S4: Generate batch correction suggestions based on the third summary item group mentioned above.

[0096] The aforementioned batch correction suggestions include: the location of the second file block and correction assistance information. The location of the second file block indicates the file position of the information file block corresponding to the third summary item in the aforementioned enterprise information file. The correction assistance information may include: the location of the third summary item within the first summary information table and a suggested prompt. The suggested prompt may be "Please perform batch correction on the above multiple summary fill content."

[0097] In practice, due to the confidence level of the third summary item, the content of the summary item can be understood as unreliable, so we can focus on guiding corrections.

[0098] Step S5: Generate batch review suggestions based on the fourth summary item group mentioned above.

[0099] The aforementioned batch review suggestions include: the location of the target summary item and review support information. The location of the target summary item indicates the position of the fourth summary item within the first summary information table. The review support information may include: the position of the fourth summary item within the first summary information table and a suggested guiding statement. The suggested guiding statement may be, "This summary item triggers an anomaly triggering rule; please adjust the summary item content according to the anomaly triggering rule."

[0100] In practice, since the fourth summary item is the summary item that triggers a specific exception triggering rule, content updates can be guided in a targeted manner based on the triggered exception triggering rule.

[0101] Step 105: Generate second suggestion information based on the second summary information table.

[0102] In some embodiments, the aforementioned implementing entity may generate second suggestion information based on the second summary information table.

[0103] The aforementioned second suggestion information represents supplementary suggestions for the summary items in the aforementioned second summary information table.

[0104] In practice, since the summary items in the second summary information table are those for which no summary items have been filled in, supplementary suggestions for supplementing the summary item content can be generated based on the second summary information table, serving as the second suggestion information. For example, the second suggestion information could be: "Summary item XX is not filled in or the filling is not recognized. Please check whether the content for this summary item exists in the enterprise information file. If not, please fill it in again."

[0105] In some optional implementations of certain embodiments, the execution entity generates second suggestion information based on the second summary information table, including: Step S1: Extract historical enterprise information files that match the second summary information table mentioned above from the target database to obtain a set of historical enterprise information files.

[0106] The historical enterprise information files are provided by enterprise terminals other than the target enterprise terminals mentioned above. The target database can be a database used to store enterprise information files submitted by enterprises that have successfully passed the project certification. Since the summary items in the second summary information table are known, they can be used as a filtering condition to select historical enterprise information files that contain the most summary item-related data from the second summary information table.

[0107] Step S2: Based on the enterprise attributes corresponding to the target enterprise terminal, select historical enterprise information files that meet the selection criteria from the above historical enterprise information file set, and use them as target enterprise information files.

[0108] Among them, the enterprise attribute can be the enterprise's sub-sector type.

[0109] In practice, historical enterprise information files that meet the filtering criteria can be selected from the aforementioned set of historical enterprise information files through enterprise attribute matching, and these files can be used as target enterprise information files. Therefore, by combining the second summary information table and enterprise attributes for dual filtering, the historical enterprise information files most relevant to the target enterprise can be obtained and used as the target enterprise information files.

[0110] Step S3: Perform data anonymization processing on the above target enterprise information file to obtain the anonymized enterprise information file.

[0111] In practice, sensitive fields in the target company's information file can be hashed or deleted to obtain an anonymized company information file.

[0112] Step S4: Based on the above-mentioned second summary information table and the above-mentioned anonymized enterprise information file, generate a summary item supplement template as the above-mentioned second suggestion information.

[0113] In practice, multiple text segments in the anonymized enterprise information file that match the summary items in the second summary information table can be located and filled into the standard template as a summary item supplement template (second suggestion information), thereby guiding the target enterprise corresponding to the target enterprise terminal to further supplement the data.

[0114] In some optional implementations of some embodiments, the above method further includes: Step S106: Generate compliance risk information based on the first summary information table and the second summary information table.

[0115] Among them, the compliance risk information characterizes the completeness of the summary information table.

[0116] In practice, the completeness score can be generated based on the proportion of summary items in the second summary information table to the total number of summary items in the summary information table. Alternatively, it can be generated based on the proportion of summary items in the first summary information table to the total number of summary items in the summary information table. Furthermore, summary items in the summary information table can have preset scores. Therefore, based on the first and second summary information tables, the score for a completed summary item is obtained, while the score for an uncompleted summary item is set to 0, thus determining the completeness score.

[0117] Step S107: Send the first recommendation information, the second recommendation information, and the compliance risk information to the target enterprise terminal.

[0118] In practice, the aforementioned first recommendation information, the aforementioned second recommendation information, and the aforementioned compliance risk information can be sent to the aforementioned target enterprise's terminal via real-time push.

[0119] Step S108: In response to successful transmission, update the business processing progress corresponding to the target enterprise terminal.

[0120] As an example, once the transaction is successfully sent, the business processing progress can be updated to "Pending further information from the enterprise".

[0121] The above embodiments of this disclosure have the following beneficial effects: The suggestion information generation method based on enterprise information semantic understanding in some embodiments of this disclosure achieves automated and highly accurate suggestion information generation. Specifically, based on the summary item feature set, the enterprise information file submitted by the target enterprise terminal is dynamically segmented to obtain an information file block set. The summary item features correspond to the summary items in the summary information table, and the information file block corresponds to at least one summary item in the summary information table. In practice, due to word limitations, it is often impossible to parse large amounts of enterprise information files at once. Conventional methods mainly use sequential segmentation, but this segmentation method may sever the contextual association between data, thus affecting feature extraction and data understanding. Therefore, this disclosure performs dynamic file segmentation by combining the summary item features corresponding to the summary items in the summary information table. This ensures the semantic integrity within the information file block and enables the association between the information file block and the summary items in the summary information table during the segmentation stage. Next, cross-semantic feature extraction is performed on each information file block in the above information file block set to generate file block semantic features corresponding to the above information file block. The file block semantic features consist of a main semantic feature and a secondary semantic feature group. In practice, although dynamic segmentation improves the semantic integrity of each information file block, since enterprise information files are a holistic description of the target enterprise terminal, certain semantic relationships still exist between the information file blocks. Therefore, this disclosure uses cross-semantic feature extraction to capture the semantic relationships between features of information file blocks, thereby improving the expressive power of features. Next, based on the aforementioned set of information file blocks and the semantic features of the corresponding file blocks, the aforementioned summary information table is pre-filled to generate a first summary information table and a second summary information table. The first summary information table represents the filled summary items in the first summary information table, and the second summary information table represents the unfilled summary items in the second summary information table. In practice, since the association between information file blocks and summary items in the summary information table is achieved during the dynamic file segmentation stage, by combining the extracted semantic features of the corresponding file blocks, automatic matching and filling of summary items in the semantic understanding dimension can be achieved. Compared with conditional judgment-based processing methods, this significantly improves matching accuracy. Furthermore, based on the aforementioned first summary information table, first suggestion information is generated, wherein the first suggestion information represents a verification suggestion for the summary items in the first summary information table. Finally, based on the aforementioned second summary information table, second suggestion information is generated, wherein the second suggestion information represents supplementary suggestions for the summary items in the second summary information table. By combining the filled-in summary items and the unfilled summary items, corresponding verification suggestions and supplementary suggestions are automatically generated.In summary, the semantic understanding of enterprise information eliminates the need to set up and maintain a large number of judgment rules, thus avoiding the extremely high maintenance costs associated with conditional judgment-based processing methods. Furthermore, the dynamic file segmentation method, which combines summary item features, not only ensures semantic integrity but also guarantees the accuracy of subsequent suggestion generation.

[0122] Further reference Figure 4 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a suggestion information generation device based on enterprise information semantic understanding. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this suggestion information generation device based on enterprise information semantic understanding can be specifically applied to various electronic devices.

[0123] like Figure 4 As shown, a suggestion information generation device 400 based on enterprise information semantic understanding in some embodiments includes: a dynamic file segmentation unit 401, a cross-semantic feature extraction unit 402, a data pre-filling unit 403, a first generation unit 404, and a second generation unit 405. The dynamic file segmentation unit 401 is configured to dynamically segment the enterprise information file submitted by the target enterprise terminal according to a set of summary item features to obtain a set of information file blocks. The summary item features correspond to summary items in a summary information table, and the information file blocks correspond to at least one summary item in the summary information table. The cross-semantic feature extraction unit 402 is configured to extract cross-semantic features from each information file block in the set of information file blocks to generate file block semantic features corresponding to the information file blocks. The file block semantic features are composed of main semantic features... The system comprises semantic feature groups; a data pre-filling unit 403 is configured to pre-fill the summary information table based on the information file block set and the semantic features of the corresponding file blocks to generate a first summary information table and a second summary information table, wherein the first summary information table represents the filled summary items in the summary information table, and the second summary information table represents the unfilled summary items in the summary information table; a first generation unit 404 is configured to generate first suggestion information based on the first summary information table, wherein the first suggestion information represents a verification suggestion for the summary items in the first summary information table; a second generation unit 405 is configured to generate second suggestion information based on the second summary information table, wherein the second suggestion information represents a supplementary suggestion for the summary items in the second summary information table.

[0124] It is understandable that the units described in the suggestion information generation device 400 based on enterprise information semantic understanding are related to the reference. Figure 1The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the suggestion information generation device 400 based on enterprise information semantic understanding and the units contained therein, and will not be repeated here.

[0125] The following is for reference. Figure 5 It shows a schematic diagram of the structure of an electronic device (e.g., a computing device) 500 suitable for implementing some embodiments of the present disclosure. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0126] like Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory 502 or a program loaded from a storage device 508 into a random access memory 503. The random access memory 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, the read-only memory 502, and the random access memory 503 are interconnected via a bus 504. An input / output interface 505 is also connected to the bus 504.

[0127] Typically, the following devices can be connected to the input / output interface 505: input devices 506 including, for example, a touchscreen, touchpad, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc.; output devices 507 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 508 including, for example, magnetic tape, hard disk, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 5 Each box shown can represent a device or multiple devices as needed.

[0128] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a read-only memory 502. When the computer program is executed by the processing device 501, it performs the functions defined above in the methods of some embodiments of this disclosure.

[0129] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0130] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0131] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: dynamically segment the enterprise information file submitted by the target enterprise terminal according to the summary item feature set, obtaining an information file block set, wherein the summary item features correspond to summary items in the summary information table, and the information file block corresponds to at least one summary item in the summary information table; perform cross-semantic feature extraction on each information file block in the aforementioned information file block set to generate file block semantic features corresponding to the aforementioned information file block, wherein the file block semantic features consist of a main semantic feature and a secondary semantic feature group; and, according to the aforementioned information file... Based on the semantic features of the file blocks corresponding to the block set and the information file blocks, the above-mentioned summary information table is pre-filled with data to generate a first summary information table and a second summary information table. The first summary information table represents the summary items that have been filled in the above summary information table, and the second summary information table represents the summary items that have not been filled in the above summary information table. Based on the first summary information table, first suggestion information is generated, which represents verification suggestions for the summary items in the first summary information table. Based on the second summary information table, second suggestion information is generated, which represents supplementary suggestions for the summary items in the second summary information table.

[0132] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0133] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0134] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0135] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for generating suggestion information based on semantic understanding of enterprise information, characterized in that, include: Based on the summary item feature set, the enterprise information file submitted by the target enterprise terminal is dynamically segmented to obtain an information file block set, wherein the summary item features correspond to the summary items in the summary information table, and the information file block corresponds to at least one summary item in the summary information table; Cross-semantic feature extraction is performed on each information file block in the information file block set to generate file block semantic features corresponding to the information file block. The file block semantic features consist of a main semantic feature and a secondary semantic feature group. The main semantic feature represents the overall semantic feature corresponding to the information file block, and the secondary semantic features in the secondary semantic feature group represent the cross-semantic features between two information file blocks adjacent to the information file block and the information file block. Based on the information file block set and the semantic features of the file blocks corresponding to the information file blocks, the summary information table is pre-filled with data to generate a first summary information table and a second summary information table. The first summary information table represents the summary items that have been filled in the summary information table, and the second summary information table represents the summary items that have not been filled in the summary information table. Based on the first summary information table, first suggestion information is generated, wherein the first suggestion information represents a verification suggestion for the summary item in the first summary information table; Based on the second summary information table, second suggestion information is generated, wherein the second suggestion information represents supplementary suggestions for summary items in the second summary information table. The step involves dynamically segmenting the enterprise information file submitted by the target enterprise terminal based on the feature set of the summary items, resulting in a set of information file blocks, including: Based on the company information documents, perform the following processing steps: Based on the preset word threshold, the enterprise information file is divided into file blocks to obtain the initial information file block. The preset word threshold represents the maximum number of words processed in a single time. The initial file block is composed of a sequence of descriptive sentences. Determine the descriptive sentence features corresponding to each descriptive sentence in the descriptive sentence sequence to obtain the descriptive sentence feature sequence. The descriptive sentence features include: descriptive sentence semantic features and semantic keyword feature group. The semantic keyword features represent the keyword features corresponding to the semantic keywords contained in the descriptive sentence. The descriptive sentence semantic features are the overall semantic expression of the descriptive sentence. The semantic keywords are the keywords in the descriptive sentence. The semantic keyword features are the semantic expressions corresponding to the keywords. For each descriptive sentence feature in the descriptive sentence feature sequence, feature matching is performed between the descriptive sentence feature and the summary item feature set to generate a matching vector corresponding to the descriptive sentence feature; Based on the obtained matching vector sequence, the description sentence sequence is filtered to obtain the filtered description sentence sequence, which is used as the information file block; The processing step ends when the enterprise information file containing the filtered descriptive sentence sequence is empty. In response to the fact that the enterprise information file after removing the filtered description sequence is not empty, the enterprise information file after removing the filtered description sequence is used as the enterprise information file, and the processing steps are performed again. The step of performing cross-semantic feature extraction on each information file block in the information file block set to generate the file block semantic features corresponding to the information file block includes: A first description sentence semantic feature sequence, a second description sentence semantic feature sequence, and a third description sentence semantic feature sequence are determined. The first description sentence semantic feature sequence is the description sentence semantic feature sequence corresponding to the information file block. The second description sentence semantic feature sequence is the description sentence semantic feature sequence corresponding to the preceding information file block. The third description sentence semantic feature sequence is the description sentence semantic feature sequence corresponding to the following information file block. The preceding information file block is located before and adjacent to the information file block, and the following information file block is located after and adjacent to the information file block. Global semantic feature extraction is performed on the first description sentence semantic feature sequence, the second description sentence semantic feature sequence, and the third description sentence semantic feature sequence to obtain the first file block semantic features, the second file block semantic features, and the third file block semantic features. The semantic features of the first file block are determined to be the main semantic features included in the semantic features of the file block corresponding to the information file block; The semantic features of the first file block and the second file block, as well as the semantic features of the first file block and the third file block, are respectively subjected to associated semantic feature extraction to obtain the semantic feature group of the file block semantic features corresponding to the information file block.

2. The suggestion information generation method based on enterprise information semantic understanding according to claim 1, characterized in that, The method further includes: Based on the first summary information table and the second summary information table, generate compliance risk information; The first suggestion information, the second suggestion information, and the compliance risk information are sent to the target enterprise terminal; Upon successful transmission, the business processing progress corresponding to the target enterprise terminal is updated.

3. The suggestion information generation method based on enterprise information semantic understanding according to claim 2, characterized in that, The step of pre-filling the summary information table with data based on the information file block set and the semantic features of the corresponding file blocks to generate a first summary information table and a second summary information table includes: For each information file block in the information file block set, perform the following prefilling steps: A first matching result is generated based on the main semantic features corresponding to the information file block and the summary item features of at least one summary item corresponding to the information file block; A second matching result is generated based on the semantic feature group corresponding to the information file block and the summary item features of at least one summary item corresponding to the information file block; The first matching result and the second matching result are correlated and verified to generate a third matching result, wherein the third matching result includes: the name of the summary item, the content of the summary item filling and the filling confidence; Based on the third matching result, at least one summary item in the summary information table corresponding to the information file block is populated with summary items; Generate the first summary information table based on the summary items already filled in the summary information table; Generate a second summary information table based on the unfilled summary items in the summary information table.

4. The suggestion information generation method based on enterprise information semantic understanding according to claim 3, characterized in that, The step of generating first suggestion information based on the first summary information table includes: The first summary information table is filtered to obtain a first summary item group, a second summary item group, a third summary item group, and a fourth summary item group. The first summary item is a summary item whose corresponding fill confidence is greater than a first confidence threshold. The second summary item is a summary item whose corresponding fill confidence is less than or equal to the first confidence threshold but greater than the second confidence threshold. The third summary item is a summary item whose corresponding fill confidence is less than or equal to the second confidence threshold. The fourth summary item is a summary item whose fill content has an anomaly with the corresponding summary item. Based on the first summary item group, generate batch confirmation suggestions; Based on the second summary item group, a batch verification suggestion is generated, wherein the batch verification suggestion includes: the location of a first file block and verification auxiliary information, wherein the location of the first file block represents the file location of the information file block corresponding to the second summary item in the enterprise information file; Based on the third summary item group, a batch correction suggestion is generated, wherein the batch correction suggestion includes: the location of the second file block and correction assistance information, and the location of the second file block represents the file location of the information file block corresponding to the third summary item in the enterprise information file; Based on the fourth summary item group, a batch review suggestion is generated, wherein the batch review suggestion includes: target summary item location and review assistance information, wherein the target summary item location represents the summary item location of the fourth summary item in the first summary information table.

5. The suggestion information generation method based on enterprise information semantic understanding according to claim 4, characterized in that, The step of generating second suggestion information based on the second summary information table includes: Historical enterprise information files that match the second summary information table are extracted from the target database to obtain a set of historical enterprise information files, wherein the historical enterprise information files are provided by enterprise terminals other than the target enterprise terminal; Based on the enterprise attributes corresponding to the target enterprise terminal, historical enterprise information files that meet the filtering conditions are selected from the historical enterprise information file set and used as target enterprise information files; The target enterprise information file is anonymized to obtain an anonymized enterprise information file; Based on the second summary information table and the anonymized enterprise information file, a summary item supplement template is generated as the second suggested information.

6. A suggestion information generation apparatus based on enterprise information semantic understanding, applied to performing the suggestion information generation method based on enterprise information semantic understanding as described in any one of claims 1 to 5, characterized in that, include: The dynamic file segmentation unit is configured to dynamically segment the enterprise information file submitted by the target enterprise terminal according to the summary item feature set to obtain an information file block set, wherein the summary item features correspond to the summary items in the summary information table, and the information file block corresponds to at least one summary item in the summary information table; The cross-semantic feature extraction unit is configured to perform cross-semantic feature extraction on each information file block in the information file block set to generate file block semantic features corresponding to the information file block, wherein the file block semantic features consist of a main semantic feature and a secondary semantic feature group; The data pre-filling unit is configured to pre-fill the summary information table according to the information file block set and the semantic features of the file blocks corresponding to the information file blocks, so as to generate a first summary information table and a second summary information table, wherein the first summary information table represents the filled summary items in the summary information table, and the second summary information table represents the unfilled summary items in the summary information table; The first generation unit is configured to generate first suggestion information based on the first summary information table, wherein the first suggestion information represents a verification suggestion for the summary item in the first summary information table; The second generation unit is configured to generate second suggestion information based on the second summary information table, wherein the second suggestion information represents supplementary suggestions for summary items in the second summary information table.

7. An electronic device, characterized in that, include: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method as described in any one of claims 1 to 5.

8. A computer-readable medium, characterized in that, It stores a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described in any one of claims 1 to 5.

Images