Table generation method, system, device and storage medium based on artificial intelligence

Abstract

The present application relates to the technical field of data processing, and specifically provides a table generation method, system, device and storage medium based on artificial intelligence, comprising: obtaining text data, and extracting entities from the text data; determining attributes of the entities through syntactic context analysis, and caching corresponding attribute values; matching pre-configured user attention items with the attributes, determining table columns according to the matched attributes, and determining table rows according to the number of entities in an adjacency matrix; generating a table based on the table columns and the table rows, and filling the attribute values of the entities into corresponding positions of the table. The table generated by the present application has complete content and strong structure adaptability, and can effectively meet the needs of multi-dimensional and customized presentation of information in different scenarios.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data processing, and particularly relates to a table generation method, system, device and storage medium based on artificial intelligence. BACKGROUND

[0002] In the prior art, automatic form filling is an important means to improve data processing efficiency. The commonly used method at present usually relies on a pre-defined structured template. The core process is: first, identifying and extracting the preset target field information from the source data (such as text, database records, etc.); then, filling the extracted target field values directly into the corresponding positions of the pre-configured table template with fixed format and table items (rows / columns).

[0003] However, this method has significant limitations. First, its function is limited to filling simple, discrete field values, such as name, date, number, etc. single information point. For more complex, hierarchical or contextually understood attribute information (such as event description, entity relationship, dynamic state, etc.) contained in the data, the existing method often has difficulty in effectively identifying and extracting.

[0004] Secondly, and most importantly, it is static and rigid. The structure of the table (including table item content, column title, number of rows, etc.) is completely determined by the pre-configured template, and cannot be dynamically adjusted according to the actual extracted data content, data size, or user-specific, temporary focus points. For example, when the source data contains additional attributes not defined in the template, these valuable information will be ignored; and when the user needs to focus on different data dimensions, the template must be redesigned and configured.

[0005] This form filling method without context understanding and dynamic adaptation ability results in incomplete and inflexible table information, which is difficult to meet the needs of multi-dimensional and customized presentation of information in complex and variable scenarios (such as government report generation, financial data analysis, medical record sorting, etc.). Therefore, there is an urgent need for a solution that can more intelligently understand data semantics and dynamically construct table structures to adapt to different data and user needs. SUMMARY

[0006] In view of the above deficiencies of the prior art, the present application provides a table generation method, system, device and storage medium based on artificial intelligence to solve the above technical problems.

[0007] In a first aspect, the present application provides a table generation method based on artificial intelligence, comprising:

[0008] obtaining text data, and extracting entities from the text data;

[0009] determine attributes of the entity through syntactic context analysis, and cache corresponding attribute values;

[0010] match the pre-configured user attention items with the attributes, determine table columns according to the matched attributes, and determine table rows according to the number of entities in the adjacency matrix;

[0011] generate a table based on the table columns and the table rows, and fill the attribute values of the entity into corresponding positions of the table.

[0012] In an optional implementation, the text data is obtained, and entities are extracted from the text data, including:

[0013] obtaining text data input by a front end;

[0014] identifying a plurality of entities from the text data by using a named entity recognition technology;

[0015] performing clustering processing on the plurality of entities, and merging entities with a similarity exceeding a set threshold into the same entity;

[0016] generating an entity list according to the merged entities.

[0017] In an optional implementation, merging entities with a similarity exceeding a set threshold into the same entity includes:

[0018] assigning a uniform concept to an entity category, and replacing entity names belonging to the entity category with the uniform concept.

[0019] In an optional implementation, determining attributes of the entity through syntactic context analysis, and caching corresponding attribute values includes:

[0020] constructing a syntactic dependency tree for the text data;

[0021] extracting subject-predicate-object triples from the syntactic dependency tree;

[0022] extracting action pairs containing entities from the subject-predicate-object triples, and predicting attributes of the entities based on the action pairs by using a BiLSTM model;

[0023] locating positions of the attributes in the text data, and identifying corresponding attribute values based on the positions.

[0024] In an optional implementation, the method further includes:

[0025] constructing a basic adjacency matrix according to the number of entities and the number of attributes;

[0026] respectively calculating co-occurrence frequencies of entities and attributes in the text data and semantic similarities of the entities and the attributes;

[0027] obtaining an edge weight of an entity and an attribute by calculating a weighted sum of the co-occurrence number and the semantic similarity;

[0028] filling the base adjacency matrix according to the edge weight of the entity and the attribute.

[0029] In an optional implementation, the method further comprises:

[0030] obtaining a keyword configured by a user, and storing the keyword as an explicit attention item;

[0031] obtaining a user operation log, the user operation log recording an operation behavior of the user on a historically generated table;

[0032] determining an implicit attention item and an attention degree of each table item according to the operation behavior;

[0033] the attention degree is used to determine a front and back position of a table item corresponding to the attribute in the table.

[0034] In an optional implementation, matching a pre-configured user attention item with an attribute in the adjacency matrix comprises:

[0035] respectively calculating a semantic similarity of the user attention item and the attribute, and marking an attribute with a semantic similarity reaching a set similarity threshold as a display attribute; the semantic similarity comprises an explicit similarity and an implicit similarity.

[0036] In a second aspect, the present application provides a table generation system based on artificial intelligence, comprising:

[0037] a first recognition module, configured to obtain text data, and extract an entity from the text data;

[0038] a second recognition module, configured to determine an attribute of the entity through syntactic context analysis, and cache a corresponding attribute value;

[0039] a relationship construction module, configured to construct an adjacency matrix of the entity and the attribute;

[0040] a parameter determination module, configured to match a pre-configured user attention item with an attribute, determine a table column according to an attribute matched successfully, and determine a table row according to a number of entities in the adjacency matrix;

[0041] a table generation module, configured to generate a table based on the table column and the table row, and fill the attribute value of the entity into a corresponding position of the table.

[0042] In a third aspect, an apparatus is provided, comprising:

[0043] a memory, configured to store a table generation program based on artificial intelligence;

[0044] a processor configured to implement the steps of the artificial intelligence based table generation method according to the first aspect when executing the artificial intelligence based table generation program.

[0045] In a fourth aspect, a computer readable storage medium is provided, and the artificial intelligence based table generation program is stored on the computer readable storage medium, and the steps of the artificial intelligence based table generation method according to the first aspect are implemented when the artificial intelligence based table generation program is executed by a processor.

[0046] The artificial intelligence based table generation method, system, device and storage medium provided by the present application have the beneficial effects that, through syntactic context analysis, the artificial intelligence based table generation method, system, device and storage medium provided by the present application can not only identify simple entities, but also accurately analyze complex attributes (such as states and relationships) of the entities, thereby significantly improving the ability to capture deep semantic information in unstructured text. Based on intelligent matching of user attention items and extracted attributes, the table column titles and the number of rows are automatically determined, and the limitation of predefined templates is completely eliminated, so that the optimal table structure can be dynamically generated according to the actual data content and user demand. The table generated by the present application has complete content and strong structure adaptability, and can effectively meet the demand for multi-dimensional and customized presentation of information in different scenarios (such as report generation and data analysis), thereby greatly improving the intelligent level and practicability of table generation. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor.

[0048] Figure 1 is a schematic flowchart of a method according to an embodiment of the present application.

[0049] Figure 2 is a schematic block diagram of a system according to an embodiment of the present application.

[0050] Figure 3 is a structural schematic diagram of a device according to an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the technical personnel in the art better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description herein is for describing particular embodiments only and is not intended to be limiting of the application.

[0053] The method for generating a table based on artificial intelligence provided by the embodiments of the application is executed by a computer device, and accordingly, the table generating system based on artificial intelligence runs in the computer device.

[0054] Figure 1 is a schematic flowchart of the method of an embodiment of the application. In this flowchart, Figure 1 The execution subject can be a table generating system based on artificial intelligence. The order of the steps in the flowchart can be changed, and some steps can be omitted according to different requirements.

[0055] As Figure 1 shown, the method comprises:

[0056] S1. obtaining text data, and extracting entities from the text data;

[0057] S2. determining attributes of the entities through syntactic context analysis, and caching corresponding attribute values;

[0058] S3. matching pre-configured user attention items with the attributes, determining table columns according to the attributes that are successfully matched, and determining table rows according to the number of entities in the adjacency matrix;

[0059] S4. generating a table based on the table columns and the table rows, and filling the attribute values of the entities into corresponding positions of the table.

[0060] In an embodiment of the application, based on step S1, a possible embodiment will be given below to non-restrictively describe the specific implementation thereof.

[0061] S101. obtaining text data input from the front end.

[0062] An event-driven asynchronous data capture mechanism is adopted to realize real-time text stream collection by listening to the HTML5Input Event interface. The data transmission frequency is limited to 50 ms / time by using the throttle technology to balance the user experience and the processing pressure of the back end. The Web Socket protocol is used to establish a bidirectional communication channel to realize segmented stream transmission of the text data, effectively reducing the transmission delay when a large amount of data is input. The AES-256 algorithm is used for end-to-end encryption during data transmission to ensure the security of user privacy data.

[0063] S102. Identify multiple entities from the text data using named entity recognition technology.

[0064] A hybrid entity recognition model was constructed, using the BERT-large pre-trained language model for word vector representation at the bottom layer and introducing a multi-head attention mechanism to capture long-range dependencies in text. During the NER tag decoding phase, a joint decoding framework was constructed by integrating the Conditional Random Field (CRF) and Bidirectional Long Short-Term Memory (BiLSTM) network to effectively address entity boundary ambiguity. For entity recognition in low-resource domains, a meta-learning-based parameter initialization method was designed, which, through 5-shot learning, can achieve the same training results as traditional methods using only 500 samples. The system supports custom entity type extensions and enables domain knowledge injection through an ontology mapping mechanism.

[0065] S103. Clustering is performed on the plurality of entities, and entities whose similarity exceeds a set threshold are grouped into entities of the same type; a unified concept is assigned to the entity category, and entity names belonging to the same entity category are replaced with the unified concept.

[0066] The entity clustering algorithm based on semantic graph embedding first constructs an entity co-occurrence network and uses the Node2Vec algorithm to generate a low-dimensional vector representation of the entity. The clustering process uses the density peak clustering (DPC) algorithm, and the cluster center and the number of clusters are automatically determined by calculating the decision diagram. The similarity calculation integrates the cosine similarity of word vectors, the similarity of hyponymy, and the similarity of distribution to construct a multidimensional similarity space. The concept merging stage adopts an ontology-driven concept mapping mechanism, constructs a concept hierarchy system through WordNet and DBpedia, and uses the OWL semantic reasoning engine to realize the mapping conversion from entity names to unified concepts. The system supports dynamic threshold adjustment and automatically determines the optimal similarity threshold through information entropy calculation.

[0067] S104. Generate an entity list based on the merged entities.

[0068] A multi-dimensional entity list generation algorithm was designed. At the data structure level, a prefix tree (Trie) was used for efficient entity retrieval, supporting entity prefix matching with O(k) time complexity. The list display employed a hierarchical visualization strategy, with multi-dimensional sorting based on entity category, confidence, and frequency of occurrence. To enhance the human-computer interaction experience, an entity importance ranking algorithm based on an attention mechanism was implemented, automatically placing key entities at the top of the display. For data persistence, a Neo4j graph database was used to store semantic relationships between entities, supporting the subsequent construction of a knowledge graph.

[0069] In an embodiment of the present invention, based on step S2, a possible embodiment will be given below to illustrate its specific implementation scheme in a non-limiting manner.

[0070] S201. Constructing a syntactic dependency tree for the text data.

[0071] A syntactic analysis model based on neural networks is used for text data processing. The model uses a pre-trained Transformer encoder (such as RoBERTa-large) as the backbone network, and captures long-distance dependency relationships between words in the text through a multi-head attention mechanism. During training, an dependency relationship prediction module based on Biaffine attention is introduced, which takes the context representation vector of each word as input, calculates the dependency relationship score matrix through a double-layer fully connected network, and then predicts the dependency relationship label between words. To improve the adaptability of the model to different language structures, a multi-language pre-training and domain self-adaptive fine-tuning strategy is used. Based on the pre-trained model in the general domain, the specific domain corpus is fine-tuned. At the same time, by introducing a syntactic constraint loss function, the learning of the model on the syntactic rules is enhanced, and the generated syntactic dependency tree is ensured to meet the language specification.

[0072] S202. Extracting subject-predicate-object triples from the syntactic dependency tree.

[0073] Based on the constructed syntactic dependency tree, a triple extraction algorithm based on the combination of rules and statistics is designed. First, according to the definition of linguistics, the subject-predicate-object extraction rule template is formulated, for example, the rule is constructed with the dependency relationship labels such as "nsubj (nominal subject)", "dobj (direct object)", and "root (sentence core verb)". For each sentence, traverse the syntactic dependency tree, and according to the rule template, match the node combination that meets the conditions to extract potential subject-predicate-object triples. To handle complex sentence patterns and semantic ambiguity, a statistical model is introduced for auxiliary verification. By statistically analyzing the common subject-predicate-object structure patterns and their co-occurrence frequencies in a large corpus, a probability model is constructed. For the candidate triples extracted by the rule, the confidence score under the statistical model is calculated, and a threshold is set to filter low-confidence results, and finally an accurate set of subject-predicate-object triples is obtained.

[0074] S203. Extracting action pairs containing entities from the subject-predicate-object triples, and predicting the attributes of the entities based on the action pairs using a BiLSTM model.

[0075] From the extracted subject-verb-object triples, action pairs (i.e., subject-verb or verb-object structures) containing the identified entity are selected. The word sequences in these action pairs are used as input for word vector representation. A pretrained word vector model (such as GloVe) is used to map each word into a low-dimensional dense vector. Positional encoding is used to incorporate the word's position in the sentence into the vector representation. This processed vector sequence is then fed into a bidirectional long short-term memory (BiLSTM) network. The BiLSTM uses forward and backward propagation to fully capture the contextual semantic information of the words in the action pair, outputting a hidden state vector at each time step. At the output layer, a fully connected layer is used to map the final hidden state vector into a probability distribution over attribute categories. The Softmax function is used to calculate the probability of each attribute category, and the category with the highest probability is selected as the predicted entity attribute. To mitigate model overfitting, Dropout regularization is introduced into the network, and an early stopping strategy is employed to optimize model training.

[0076] S204. Locate the position of the attribute in the text data, and identify the corresponding attribute value based on the position.

[0077] After predicting the entity attributes, the attribute values ​​are located based on the syntactic dependency tree and the original position information of the text. First, according to the predicted attribute category, the grammatical structure path related to the attribute is searched in the syntactic dependency tree. For example, for descriptive attributes, they are usually searched in structures modified by adjectives or adverbs; for relational attributes, they are located based on related prepositional phrases or clause structures. Then, combined with the character offset information of the text, the starting and ending positions of the attribute value are determined in the original text data. To improve the positioning accuracy, the conditional random field (CRF) model is introduced to optimize the preliminary positioning results. The text features such as word vectors, part-of-speech tags, and syntactic dependency relationship tags are used as inputs to the CRF. By learning the transition probability and feature function between labels, the most likely attribute value boundary label sequence is predicted, thereby accurately identifying the attribute value.

[0078] In one embodiment of the present invention, the relationship between entities and attributes may be stored as an adjacency matrix.

[0079] 1. Construct a basic adjacency matrix based on the number of entities and the number of attributes.

[0080] Let the entity set be E={e1,e2,…,e m}, the attribute set is A={a1,a2,…,a n}, construct the m×n dimensional basic adjacency matrix The row index of the matrix corresponds to the entity, the column index corresponds to the attribute, and the initial value of all elements is 0, that is, In order to optimize storage efficiency, a sparse matrix storage structure is used for large-scale entity attribute networks, and the proportion of non-zero elements is .

[0081] 2. Calculate the co-occurrence frequency and semantic similarity of entities and attributes in text data respectively.

[0082] Design a sliding window mechanism to calculate the co-occurrence frequency of entities and attributes. For a text sequence T = {t1, t2, …, t L} of length L, set the window size w (w = 10 in the experiment), and count the number of times C i e j and a i,j appear simultaneously in the window. Adopt TF-IDF weighting strategy to enhance key co-occurrence relationship, and define the weighted co-occurrence frequency as:

[0083]

[0084] Where |D| is the total number of corpus documents, and |d:e i ,a j ∈d| is the number of documents containing entity e i and attribute a j .

[0085] Adopt hybrid embedding model to calculate semantic similarity. First, get the context embedding vector of entity and attribute through BERT-large, and then combine the structured vector generated by the domain knowledge graph embedding (such as TransE) to construct the fusion representation:

[0086]

[0087] Where α = 0.7 is a hyperparameter. The semantic similarity is defined as:

[0088] .

[0089] 3. Calculate the edge weight of entities and attributes by calculating the weighted sum of the co-occurrence frequency and the semantic similarity.

[0090] Propose an adaptive weighted fusion algorithm to calculate the edge weight. Define the weight function:

[0091]

[0092] Where λ ∈ [0, 1] is the adaptive weight parameter, which is solved by the following optimization problem:

[0093]

[0094] Calibrate the parameters using the gold standard dataset, and finally get the optimal weight λ * = 0.62.

[0095] 4. Fill the base adjacency matrix according to the edge weight of entity and attribute.

[0096] A parallel filling algorithm is designed to improve the efficiency of matrix construction. Distributed computing is realized by using the MapReduce framework, and entity-attribute pairs are divided into p data blocks, and each computing node processes m / p rows of entity data. In the filling process, a threshold filtering mechanism is introduced, and when W{i,j} < θ (when θ = 0.2, M i,j =0 is set to reduce noise connections. Finally, the weighted adjacency matrix M is obtained, and the element distribution conforms to the power-law distribution characteristics, that is, .

[0097] By constructing the adjacency matrix of entities and attributes, unstructured text is converted into a computable relationship network, and the adjacency matrix can be used as a knowledge storage structure evolving over time.

[0098] In an embodiment of the present application, based on step S3, a possible embodiment will be given below to non-restrictively illustrate the specific implementation thereof.

[0099] Firstly, the user attention items are determined, and the specific method comprises:

[0100] (1) Obtain the keywords configured by the user, and store the keywords as explicit attention items.

[0101] An explicit attention item management framework based on semantic expansion is constructed to realize accurate capture and structured storage of user demand. The system receives the original keyword set configured by the user through an interactive interface, and uses a three-layer semantic expansion strategy to improve the semantic coverage of the keywords: firstly, synonym expansion is performed based on semantic resources such as WordNet to solve the problem of user expression diversity; secondly, the upper and lower concepts of the keywords are obtained through domain ontology construction technology to form a hierarchical semantic network; finally, the semantic similarity of word vectors is calculated by using a pre-trained language model (such as BERT) to mine implicit semantic associated words. The expanded keyword set is stored in an inverted index structure to construct a "keyword-document" mapping table, supporting fast retrieval with O(logn) time complexity. In order to adapt to the dynamic update requirement, an incremental index update mechanism is designed, and when the user adds or deletes keywords, the local update of the index is triggered through an event-driven mode to ensure the real-time and high efficiency of the storage structure.

[0102] (2) Obtain the user operation log, which records the user's operation behavior on the historical generated table.

[0103] The five-tuple structure records the user's interaction behavior with the table, including user identification, table identification, operation type, timestamp, and operation duration. The operation type covers 12 basic interaction behaviors such as viewing, filtering, sorting, and exporting, each of which is assigned a different semantic weight to reflect the user's attention intensity. The log preprocessing stage adopts a sliding time window mechanism to focus on recent behavior patterns, and the window size is dynamically adjusted according to the stability of user behavior (the default setting is 7 days). To reflect the timeliness of behavior, an exponential decay model is introduced to weight the historical operations, and the operations closer to the current time are assigned higher weight coefficients. At the same time, a log cleaning mechanism is implemented to filter abnormal operation records (such as invalid operations with a duration of less than 50ms), and a distributed message queue (such as Kafka) is used to realize real-time collection and transmission of logs, ensuring data integrity and low latency characteristics.

[0104] (3) determining implicit attention items and attention degrees of each table item according to the operation behavior; the attention degree is used to determine the front and back positions of the table item corresponding to the determined attribute in the table.

[0105] A user-table-table item three-layer heterogeneous graph model is constructed. The nodes in the graph model include three types of entities: users, tables, and table items. The edge relationships include the interaction edges between users and tables (the weight is determined by the operation behavior intensity) and the containing edges between tables and table items (the weight is a fixed value). The graph attention network (GAT) architecture is used to calculate the attention degree of the table item, and the multi-head attention mechanism is used to automatically learn the correlation weight between different types of nodes. The attention degree calculation integrates multi-dimensional features of operation behavior: operation duration reflects the cognitive cost of user investment, operation frequency reflects the persistence of attention, and operation recency dynamically adjusts the weight through a time decay function. To improve the generalization ability of the model, a transfer learning strategy is introduced, which fine-tunes the user behavior data in a specific domain based on a pre-trained model in a general domain.

[0106] Then, according to the user attention item, the attributes and table parameters that need to be displayed are determined:

[0107] S301. Calculate the semantic similarity of the user attention item and the attribute respectively, and mark the attribute with a semantic similarity that reaches a set similarity threshold as a display attribute; the semantic similarity includes explicit similarity and implicit similarity.

[0108] Explicit Similarity Calculation Mechanism: Utilizing a word embedding model trained on a large-scale corpus, the system calculates the semantic distance between the keyword and the attribute name. For example, when a user searches for "smartphone," the system identifies the semantic relevance of attributes like "processor model" and "screen resolution" to the keyword. Domain knowledge graphs are introduced to construct a semantic association network between entities. For example, the knowledge graph shows that "camera pixels" is a sub-attribute of "photographic function." Therefore, when a user searches for "photography performance," the system automatically associates the keyword with the relevant attributes.

[0109] Implicit similarity calculation mechanism: Record user operations on the table (such as filtering, sorting, and exporting) and analyze which attributes are frequently used. For example, if a user frequently filters by "price range," this indicates a high correlation between this attribute and the user's preferred items. Analyze the co-occurrence frequency of the user's preferred items and attributes in historical operations. For example, if a user frequently searches for "graphics card model" and "memory capacity" while browsing content related to "laptops," this indicates an implicit correlation between these two attributes and the user's preferred items.

[0110] Hybrid Similarity Fusion Strategy: Confidence scores for explicit and implicit similarities are calculated separately. For example, when the keyword match is clear and there is a strong correlation in the knowledge graph, the confidence of explicit similarity is higher; when the user has recently frequently operated on a certain attribute, the confidence of implicit similarity is higher. The fusion weight is dynamically adjusted based on the confidence score. For example, for new users or cold start scenarios, the explicit similarity weight is higher; for old users with rich behavioral data, the implicit similarity weight is increased accordingly. Based on the distribution characteristics of the similarity scores, the density peak clustering algorithm is used to automatically identify the appropriate similarity threshold. This threshold classifies attributes into three categories: high correlation, medium correlation, and low correlation.

[0111] Display attribute marking rule: Attributes whose similarity scores exceed the threshold are directly marked as display attributes.

[0112] S302. Determine table columns according to display attributes, and determine table rows according to the number of entities in the adjacency matrix.

[0113] Identify table columns:

[0114] Receives a list of all attributes after entity resolution, along with the user's attention weight for each attribute. User attention weights can be obtained through explicit user settings (such as keyword configuration) or through system analysis of user implicit behavior (such as operation logs).

[0115] Sorting based on attention: First, the final weight of each attribute is calculated. This weight is the fusion result of the user's explicit weight and the implicit weight calculated by the system. The user's explicit weight is directly taken from the user's settings. If the user does not set it, the default value of 0.5 is used.

[0116] The formula for calculating the most important weight is: most important weight = 0.7 * explicit weight + 0.3 * implicit weight. In this way, both the user's explicit needs and potential interests are considered.

[0117] To reflect the importance of the user's recent behavior, the system also introduces a time decay factor. For operations that are far from the current time, the corresponding attribute weight will decay over time. The decay formula is: decayed weight = original weight * e^(-0.1 * past days).

[0118] The sorted attributes are processed in layers, and they are divided into different priorities according to their attention. The top 3 attributes with the highest attention are selected as the main column, with a priority of 1. The main column is usually the core attribute of the entity, such as name, price, model, etc. These attributes will be prominently displayed in the table. The next 3 attributes with medium attention are selected as the secondary column, with a priority of 2. The secondary column provides supplementary information, such as rating, sales, size, etc. Their display position will be inferior to the main column. In addition to the main column and the secondary column, the system will also add a "details" column, which is an expandable panel. This column is used to contain all the remaining attributes to avoid layout confusion caused by too many table columns.

[0119] Adaptive column width calculation: Calculate the appropriate width for each column to ensure that the table content is displayed clearly and aesthetically. The calculation of column width is based on two factors: the length of the attribute name and the maximum length of the attribute value. The base width is obtained by multiplying the length of the attribute name by 8 pixels, and the maximum content width is obtained by multiplying the maximum length of the attribute value by 6 pixels. The final column width takes the larger value of the base width and the maximum content width, but limits the maximum width to 200 pixels to prevent overly wide columns from affecting the overall layout of the table.

[0120] Determine the table row:

[0121] Receive the table data after entity resolution processing, and extract all unique entities. These entities may exist in different forms of expression, but the system needs to identify whether they belong to the same actual object.

[0122] Analyze all extracted entities, first determine which entities belong to the same category based on semantic similarity. For example, "iPhone14" and "Apple phone", calculate their semantic similarity through a pre-trained language model, and when the similarity exceeds a certain threshold (such as 0.8), they are determined as the same category of entity.

[0123] In addition to semantic similarity, the system also performs entity merging based on domain-specific rules. In the electronics product domain, different expressions such as "iPhone14Pro" and "Apple14Pro" are unified into the standard expression "iPhone14Pro" according to pre-set rules.

[0124] Generating entity unique ID: MurmurHash3 algorithm is used to generate the unique ID of the entity, which has the characteristics of high efficiency and low collision rate. The input of generating ID is the string obtained by splicing the entity type (such as "electronic product") and the core attribute (usually the combination of brand and model). In this way, the same class of entities can generate the same ID. When there is a hash collision, the system will use a chained hash table to solve the conflict, and add an incremental serial number after the ID to ensure the uniqueness of each entity ID.

[0125] In an embodiment of the present application, based on step S4, a possible embodiment will be given below to non-limitingly illustrate the specific implementation thereof.

[0126] Initialize the table structure according to the results of column construction and row construction:

[0127] Create table container: generate standard HTML table framework, including

[0128]

[0129]

[0130]

[0131]

[0132]

[0133]

[0134]

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142] Figure 1

[0143] Figure 2

[0144]

[0145]

[0146]

[0147]

[0148] Figure 3

[0149]

[0150]

[0151]

[0152]

[0153]

[0154]

[0155]

[0156]

[0157]

[0158]

[0159] 、 and Label; header generation: generate table header cells based on column configuration, set column width, alignment, etc. properties; row frame generation: create table rows for each entity, pre-allocate row height and other style properties. Property value mapping and filling: property matching rule: exact match based on property name (e.g. "name" corresponds to "name" column); support alias mapping configuration (e.g. "price" mapped to "price" column); implement fuzzy matching mechanism (e.g. properties containing "model" keyword mapped to "model" column). Data type processing: numerical type: apply thousandth separator, retain appropriate decimal places; date type: format according to user preference (e.g. YYYY-MM-DD); boolean type: converted to "yes / no" or represented using icons; array type: connected using specific separators (e.g. comma, vertical bar). Value conversion and formatting: currency formatting: add currency symbol according to region setting; percentage formatting: automatically add percentage sign and retain two decimal places; unit conversion: automatically convert weight units to commonly used units; text truncation: automatically truncate long text and add ellipsis. In some embodiments, the artificial intelligence-based table generation system can include a plurality of functional modules composed of computer program segments. The computer programs of each program segment in the artificial intelligence-based table generation system can be stored in the memory of the computer device and executed by at least one processor to perform the functions of artificial intelligence-based table generation (see description for details). In this embodiment, the artificial intelligence-based table generation system can be divided into a plurality of functional modules according to the functions it performs, as shown. The module referred to in the present application refers to a series of computer program segments that can be executed by at least one processor and can complete a fixed function, which are stored in the memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments. A first identification module is configured to obtain text data and extract entities from the text data; a second identification module is configured to determine the attributes of the entities through syntactic context analysis and cache the corresponding attribute values; a parameter determination module is configured to match pre-configured user attention items with attributes, determine table columns according to the attributes that match successfully, and determine table rows according to the number of entities in the adjacency matrix; and a table generation module is configured to generate a table based on the table columns and the table rows, and fill the attribute values of the entities into the corresponding positions of the table. The artificial intelligence-based table generation method provided in the embodiments of the present application can be applied to a device. Those skilled in the art can understand that the device structure involved in the embodiments of the present application does not constitute a limitation on the device, and the device can include more or fewer components than the illustration, or combine certain components, or different component arrangements. In the embodiments of the present application, the device includes but is not limited to a laptop computer, a desktop computer, a workstation, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers.The device can also represent various forms of mobile devices such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit implementations of the applications described and / or claimed in this document. The device 300 can include a processor 310, a memory 320, and a communication unit 330. These components can communicate with each other through one or more buses. As one skilled in the art can understand, the structure of the server shown in the figure is not a limitation on the present application, and it can be a bus structure or a star structure. It can also include more or fewer components than shown in the figure, or combine certain components, or have different arrangements of components. The memory 320 can be used to store instructions for execution by the processor 310. The memory 320 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the device 300 is enabled to perform some or all of the steps in the following method embodiments. The processor 310 is the control center of the storage device, which connects all parts of the electronic device through various interfaces and lines, and performs various functions of the electronic device and / or processes data by running or executing software programs and / or modules stored in the memory 320 and calling data stored in the memory. The processor can be composed of integrated circuits (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same function or different functions connected. For example, the processor 310 can only include a central processing unit (CPU). In the present application, the CPU can be a single operation core or can include multiple operation cores. The communication unit 330 is used to establish a communication channel so that the storage device can communicate with other devices. It receives user data sent by other devices or sends user data to other devices. The present application also provides a computer storage medium, which can store a program that performs some or all of the steps in the embodiments provided by the present application when executed. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present application can be implemented by means of software plus necessary universal hardware platforms. Based on such an understanding, the technical solutions in the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and the like, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a second device, a network device, or the like) to execute all or part of the steps of the methods described in the embodiments of the present application. The same or similar parts in the embodiments of the present application can be referred to each other. In particular, for the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments. In the several embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only schematic. The division of the modules is only a logical function division, and other division manners can be adopted in actual implementation, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the modules shown or discussed can be indirect coupling or communication connection through some interfaces, and can be electrical, mechanical or other forms. The modules shown as separate components can be or can not be physically separated, and the components shown as modules can be or can not be physical modules, that is, they can be located in one place, or can be distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments. In addition, the functional modules in the embodiments of the present application can be integrated in a processing module, or each module can be physically present, or two or more modules can be integrated in one module. Although the present application has been described in detail with reference to the preferred embodiments, the present application is not limited to this. Without departing from the spirit and essence of the present application, those skilled in the art can make various equivalent modifications or replacements to the embodiments of the present application, and these modifications or replacements should be included in the scope of the present application.

Claims

1. An artificial intelligence-based table generation method, characterized by, The method comprises the following steps: acquiring text data and extracting entities from the text data; determining attributes of the entities through syntactic context analysis and caching corresponding attribute values; matching pre-configured user attention items with the attributes, determining table columns according to the attributes that are successfully matched, and determining table rows according to the number of entities in an adjacency matrix; generating a table based on the table columns and the table rows, and filling attribute values of the entities into corresponding positions of the table; the method further comprises the following steps: constructing a basic adjacency matrix according to the number of entities and the number of attributes; calculating co-occurrence frequencies of entities and attributes in the text data and semantic similarities between the entities and the attributes, respectively; obtaining edge weights of the entities and the attributes by calculating a weighted sum of the co-occurrence frequencies and the semantic similarities; and filling the basic adjacency matrix according to the edge weights of the entities and the attributes; the method further comprises the following steps: acquiring user-configured keywords and storing the keywords as explicit attention items; acquiring user operation logs, which record user operation behaviors on a table that is generated in history; determining implicit attention items and attention degrees of each table item according to the operation behaviors; and the attention degrees are used to determine the positions of table items corresponding to the attributes in the table; matching pre-configured user attention items with attributes in an adjacency matrix comprises the following steps: calculating semantic similarities between the user attention items and the attributes, respectively, and marking attributes with a semantic similarity reaching a set similarity threshold as display attributes; the semantic similarities include explicit similarities and implicit similarities.

2. The method of claim 1, wherein, The method of acquiring text data and extracting entities from the text data comprises the following steps: acquiring text data input by a front end; identifying a plurality of entities from the text data by using a named entity recognition technology; performing clustering processing on the plurality of entities, and merging entities with a similarity exceeding a set threshold into the same entity; generating an entity list according to the merged entities.

3. The method of claim 2, wherein, The method of merging entities with a similarity exceeding a set threshold into the same entity comprises the following steps: allocating a unified concept to an entity category, and replacing entity names belonging to the entity category with the unified concept.

4. The method of claim 1, wherein, The method of determining attributes of the entities through syntactic context analysis and caching corresponding attribute values comprises the following steps: constructing a syntactic dependency tree for the text data; extracting subject-predicate-object triples from the syntactic dependency tree; extracting action pairs containing entities from the subject-predicate-object triples, and predicting attributes of the entities based on the action pairs by using a BiLSTM model; locating positions of the attributes in the text data, and identifying corresponding attribute values based on the positions. 5.A table generation system based on artificial intelligence, characterized by, The method comprises the following steps: a first identification module is configured to acquire text data and extract entities from the text data; a second identification module is configured to determine attributes of the entities through syntactic context analysis and cache corresponding attribute values; a parameter determination module is configured to match pre-configured user attention items with the attributes, determine table columns according to the attributes that are successfully matched, and determine table rows according to the number of entities in an adjacency matrix; a table generation module is configured to generate a table based on the table columns and the table rows, and fill attribute values of the entities into corresponding positions of the table; Also comprising: constructing a basic adjacency matrix according to the number of entities and the number of attributes; calculating the co-occurrence frequency of entities and attributes in the text data and the semantic similarity of entities and attributes respectively; obtaining the edge weight of entities and attributes by calculating the weighted sum of the co-occurrence frequency and the semantic similarity; filling the basic adjacency matrix according to the edge weight of entities and attributes; Also comprising: obtaining the keywords configured by the user and storing the keywords as explicit attention items; obtaining the user operation log, which records the operation behavior of the user on the historically generated table; determining the implicit attention items and the attention degree of each table item according to the operation behavior; the attention degree is used to determine the front and back positions of the table item corresponding to the attribute in the table; Matching the pre-configured user attention items with the attributes in the adjacency matrix, comprising: Respectively calculating the semantic similarity of the user attention items and the attributes, and marking the attributes with a semantic similarity reaching a set similarity threshold as display attributes; the semantic similarity includes explicit similarity and implicit similarity. 6.A table generating device based on artificial intelligence, characterized by Comprising: A memory for storing an artificial intelligence-based table generation program; A processor for implementing the steps of the artificial intelligence-based table generation method according to any one of claims 1-4 when executing the artificial intelligence-based table generation program.

7. A computer readable storage medium storing a computer program, characterized in that, The readable storage medium has an artificial intelligence-based table generation program stored thereon, and the artificial intelligence-based table generation program implements the steps of the artificial intelligence-based table generation method according to any one of claims 1-4 when executed by a processor.

Images