Financial data retrieval method and system based on artificial intelligence

Abstract

The invention provides a financial data retrieval method and system based on artificial intelligence. The method comprises the steps of obtaining original query data and associated information data input by a user; inputting the associated information data into the user portrait model to obtain a user category output by the user portrait model; inputting the user category and the original query data into an intention recognition model to obtain a multi-layer retrieval intention output by the intention recognition model; determining a domain knowledge graph based on the user category, and disambiguating terms in the multi-layer retrieval intention based on the domain knowledge graph to obtain a term disambiguation result; based on the term disambiguation result, reconstructing a query retrieval logic tree, and based on the query retrieval logic tree, determining a preliminary retrieval result; and performing fusion sorting on the preliminary retrieval results based on the multi-layer retrieval intention to obtain fusion retrieval results, and screening the fusion retrieval results to determine a target retrieval result. According to the invention, the speed and quality of obtaining effective information by a user are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based financial data retrieval method and system thereof. Background Art

[0002] With the rapid development of artificial intelligence (AI), more and more traditional industries are exploring and applying these advanced technologies to improve efficiency and reduce errors. In the field of financial management, AI-based financial data retrieval methods are becoming a new trend. By leveraging natural language processing and machine learning, they can quickly and accurately extract the required data from massive amounts of financial information, not only increasing data processing speed but also significantly enhancing decision support capabilities.

[0003] Although artificial intelligence (AI)-based financial data retrieval methods have shown great potential, several pressing challenges remain. First, financial data retrieval users can be divided into professionals and non-professionals. Due to significant differences in knowledge and work requirements, the two groups have fundamentally different search objectives and problem focus. This difference leads to limitations in the application of natural language processing technology in financial data retrieval. On the one hand, financial terminology is ambiguous and industry-specific, and its semantics are often influenced by context, making it difficult for AI models to accurately understand it. On the other hand, users' diverse expression habits and ambiguous needs increase the difficulty of semantic parsing, making it difficult for models to accurately match users' search intent. Furthermore, the relationship between search content and search intent is complex. Although different users may search for overlapping content, their search intent often differs. Current technologies struggle to accurately distinguish subtle differences in intent, resulting in repeated searches. This wastes computing resources, reduces search efficiency, and affects the speed and quality of users' access to effective information. Summary of the Invention

[0004] The present invention provides an artificial intelligence-based financial data retrieval method and system thereof, which are used to ensure that different types of users can obtain target retrieval results that meet their own needs, thereby improving the speed and quality of users' acquisition of effective information.

[0005] In a first aspect, the present invention provides an artificial intelligence-based financial data retrieval method, comprising:

[0006] Obtaining original query data and associated information data input by the user; inputting the associated information data into a user portrait model to obtain the user category output by the user portrait model; the user portrait model is trained based on sample associated information and its corresponding user category label results;

[0007] Inputting the user category and the original query data into an intent recognition model to obtain a multi-layered retrieval intent output by the intent recognition model; the multi-layered retrieval intent includes a main retrieval intent and a sub-retrieval intent; the intent recognition model is trained based on sample user categories, sample query data, and their corresponding retrieval intent label results;

[0008] Based on the user category, a domain knowledge graph is determined, and terms in the multi-layer search intent are disambiguated based on the domain knowledge graph to obtain a term disambiguation result;

[0009] reconstructing a query retrieval logic tree based on the term disambiguation result, and determining a preliminary retrieval result based on the query retrieval logic tree;

[0010] The preliminary search results are fused and sorted based on the multi-layer search intent to obtain a fused search result, and the fused search result is screened based on the user category to determine the target search result.

[0011] In a second aspect, the present invention further provides an artificial intelligence-based financial data retrieval system, which is applied to the artificial intelligence-based financial data retrieval method as described in the first aspect; the artificial intelligence-based financial data retrieval system comprises:

[0012] A user profile building module is used to obtain the original query data and associated information data input by the user; and input the associated information data into the user profile model to obtain the user category output by the user profile model; the user profile model is trained based on the sample associated information and its corresponding user category label results;

[0013] An intent recognition module is configured to input the user category and the original query data into an intent recognition model to obtain a multi-layered retrieval intent output by the intent recognition model; the multi-layered retrieval intent includes a main retrieval intent and sub-retrieval intents; the intent recognition model is trained based on sample user categories, sample query data, and their corresponding retrieval intent label results;

[0014] A term disambiguation module is used to determine a domain knowledge graph based on user categories, and to disambiguate terms in the multi-layer search intent based on the domain knowledge graph to obtain a term disambiguation result;

[0015] a reconstructing retrieval module, configured to reconstruct a query retrieval logic tree based on the term disambiguation result, and determine a preliminary retrieval result based on the query retrieval logic tree;

[0016] The retrieval screening module is used to fuse and sort the preliminary retrieval results based on multi-layer retrieval intentions to obtain fused retrieval results, and to screen the fused retrieval results based on user categories to determine target retrieval results.

[0017] In a third aspect, the present invention further provides an electronic device comprising: a memory for storing a computer software program; and a processor for reading and executing the computer software program, thereby implementing any one of the above-mentioned artificial intelligence-based financial data retrieval methods.

[0018] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, wherein the storage medium stores a computer software program, and when the computer software program is executed by a processor, it implements any of the artificial intelligence-based financial data retrieval methods described above.

[0019] In a fifth aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described artificial intelligence-based financial data retrieval methods.

[0020] The artificial intelligence-based financial data retrieval method provided by the embodiment of the present invention constructs user portraits by acquiring related information data, and can use different knowledge graphs for term disambiguation for different user categories, effectively solving the semantic understanding problems caused by the ambiguity and specificity of financial terms and the diversity of user expressions, and improving the accuracy of natural language processing in financial data retrieval; in addition, multi-layer retrieval intentions are derived through the intention recognition model, and the retrieval results are integrated and sorted based on this, which can accurately distinguish the different intentions behind similar retrieval content of different users, avoid repeated retrieval, improve retrieval efficiency, and reduce the waste of computing resources; finally, from user portrait construction, intention recognition, term disambiguation to retrieval result processing, and combined with user categories for personalized processing, it ensures that different types of users can obtain target retrieval results that meet their own needs, thereby improving the speed and quality of users obtaining effective information. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 1 is a flow chart of a financial data retrieval method based on artificial intelligence provided by an embodiment of the present invention;

[0022] Figure 2 1 is a schematic diagram of the structure of an artificial intelligence-based financial data retrieval system provided by an embodiment of the present invention;

[0023] Figure 3 An embodiment diagram of an electronic device provided by an embodiment of the present invention;

[0024] Figure 4 An embodiment diagram of a computer-readable storage medium provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0026] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the specified features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0027] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment of the present invention described as "for example" is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed herein.

[0028] See Figure 1 , Figure 1 : is a flow chart of the artificial intelligence-based financial data retrieval method provided by the present invention. In the embodiment of the present invention, the execution subject of the artificial intelligence-based financial data retrieval method is a financial data retrieval system. Therefore, the artificial intelligence-based financial data retrieval method includes:

[0029] Step 10, obtain the original query data and associated information data input by the user; and input the associated information data into the user portrait model to obtain the user category output by the user portrait model; the user portrait model is trained based on the sample associated information and its corresponding user category label results.

[0030] Optionally, the financial data retrieval system first obtains raw query data directly input by the user, including search keywords or phrases. It then collects related information data, including user identity tags, historical search records, and the current operation context. Specifically, based on the user identity tag, such as user ID (financial analyst or sales manager) and its corresponding identity permissions, the system retrieves historical search records (frequent searches for sales cost-related data in the past three months; upon logging into the retrieval system, the system records the user's position information and retrieves records of past queries based on this position information) and the current operation context (where the position information is the current job, such as preparing for a quarterly financial report). After obtaining the related information data, it is input into a pre-trained user profile model. This model uses supervised learning based on a large amount of sample related information and its corresponding user category labels (e.g., professional financial personnel, business department personnel, ordinary company employee, etc.), employing model architectures such as machine learning, reinforcement learning, and deep neural networks. By analyzing the occupational information in the user identity tag, the data type preferences in the historical search records, and the task requirements in the current operation context, the model outputs a user category, such as determining that the user is a "business department personnel."

[0031] In one embodiment, let's take user A as an example. The original query data is "View sales collection status for the North China region in the first quarter of 2024." The associated information data shows that their user identity tag is "Regional Sales Manager." Their historical search records focus on data such as regional sales performance and customer collections. The current operation scenario is preparing for a regional sales performance review meeting. Based on this information, the user profile model categorizes user A as a "Business Department Personnel."

[0032] In step 20, the user category and the original query data are input into the intent recognition model to obtain the multi-layer retrieval intent output by the intent recognition model; the multi-layer retrieval intent includes the main retrieval intent and the sub-retrieval intent; the intent recognition model is trained based on the sample user category, the sample query data and the corresponding retrieval intent label results.

[0033] Optionally, after obtaining a specific user category, the financial data retrieval system inputs the user category and raw query data into an intent recognition model. This model is trained based on sample user categories, sample query data, and their corresponding retrieval intent label results. The model utilizes deep learning and other artificial intelligence architectures, such as Transformer. The model first performs word segmentation and word vector embedding on the raw query data. Combined with user category information, it uses a multi-layer attention mechanism to identify multi-layered retrieval intents. Specifically, multi-layered retrieval intents include primary retrieval intents and sub-retrieval intents. These multi-layered intents refine user needs, avoid ambiguous searches, and allow subsequent searches to more accurately match user needs, improving the targeted nature of searches.

[0034] In one embodiment, continuing with the example of User A above, the intent recognition model is fed with the "Business Department Personnel" category and the original query data "View sales collection status in North China for the first quarter of 2024." Since User A is a regional sales manager, the model identifies the primary search intent as "Evaluate sales collection performance in North China," with sub-search intents including "Compare target collection amounts with actual collection amounts" and "Analyze collection delays for each customer."

[0035] Step 30: Based on the user category, a domain knowledge graph is determined, and the terms in the multi-layer search intent are disambiguated based on the domain knowledge graph to obtain a term disambiguation result.

[0036] Optionally, the financial data retrieval system first pre-sets a knowledge graph library, which contains domain knowledge graphs under corresponding user categories. Then, based on the user category determined in step 20, the corresponding domain knowledge graph is selected from the knowledge graph library. For example, if the user is a "financial auditor", a professional financial knowledge graph containing detailed accounting standards and audit specifications is selected; if the user is a "business department staff", a knowledge graph focusing on business terminology explanations is selected. Afterwards, since financial domain terms have different meanings in different scenarios and user groups, ambiguity can easily arise from literal understanding. Therefore, based on the determined domain knowledge graph, the nodes and relationships in the knowledge graph are used to eliminate ambiguous terms in multi-layer retrieval intents, as described in steps 301-305. For example, for the term "travel expenses", the definitions of the expense details, reimbursement standards, etc. included in the professional financial knowledge graph are clarified to eliminate ambiguity and obtain the accurate meaning of the term, that is, the term disambiguation result. This effectively solves the problems of ambiguity and specificity of financial terms.

[0037] In one embodiment, for user A (business department staff), the term "sales collection" appears in his multi-layer search intent. "Sales collection" is matched from the business domain knowledge graph to refer to the actual amount of sales payment paid by the customer. It is clear that it does not include information such as uncollected accounts receivable, thereby completing term disambiguation.

[0038] Step 40: reconstruct the query retrieval logic tree based on the term disambiguation result, and determine the preliminary retrieval result based on the query retrieval logic tree.

[0039] Optionally, based on the term disambiguation results, the financial data retrieval system uses a logical tree construction approach to convert each semantic unit or statement in the term disambiguation results into a structured query retrieval logical tree, as described in steps 401 through 403. The logical tree is then converted into a database query statement, and the query is executed in the financial data system to obtain preliminary search results. This converts the user's search intent into executable database query logic, improving retrieval accuracy and efficiency.

[0040] In one embodiment, for user A, a query and retrieval logic tree is constructed based on the primary and sub-search intents of "Evaluate sales collection performance in North China." The root node is "Evaluate sales collection performance in North China, Q1 2024," and branch nodes include "Actual Collection Amount Query," "Target Collection Amount Query," and "Calculate Customer Collection Delay Days." The logic tree is then converted into an SQL query statement and executed against the company's financial database to obtain preliminary search results containing fields such as the customer collection amount and collection time.

[0041] Step 50: The preliminary search results are fused and sorted based on the multi-layer search intent to obtain a fused search result, and the fused search result is screened based on the user category to determine the target search result.

[0042] Optionally, based on the preliminary search results obtained in step 40, the financial data retrieval system performs a fusion sorting of the multiple relevant data items according to the multi-layered search intent obtained in step 20. For example, weights may be assigned based on factors such as the relevance and importance of the search intent and the result data. A score is then calculated for each result, and the results are sorted according to the scores to obtain a fusion search result. Through fusion sorting, the results are filtered and sorted based on the importance of the search intent, highlighting data relevant to the user's core needs. This is described in detail in steps 501-504. Subsequently, based on the sorted fusion search results and the user category determined in step 10, the fusion search results are further filtered. Because different user categories have different data needs and sensitivities, for example, corporate executives may be more interested in summary data, while financial analysts may require detailed, detailed data, filtering can remove data that does not meet the user's knowledge level and business needs, ensuring that the final target search results are both accurate and consistent with user usage habits, ultimately obtaining the target search results. It is important to note that during filtering, filtering rules can be set based on user category. For example, for "business department personnel," overly specialized financial analysis data can be removed, while intuitive business indicator data can be retained.

[0043] The embodiment of the present invention constructs a user profile by acquiring related information data, and can use different knowledge graphs for term disambiguation for different user categories, effectively solving the semantic understanding problems caused by the ambiguity and specificity of financial terms and the diversity of user expressions, and improving the accuracy of natural language processing in financial data retrieval; in addition, through the intention recognition model, multi-layer retrieval intentions are derived, and the retrieval results are integrated and sorted based on this, which can accurately distinguish the different intentions behind similar retrieval content of different users, avoid repeated retrieval, improve retrieval efficiency, and reduce the waste of computing resources; finally, from user portrait construction, intention recognition, term disambiguation to retrieval result processing, and combined with user category for personalized processing, it ensures that different types of users can obtain target retrieval results that meet their own needs, thereby improving the speed and quality of users' acquisition of effective information.

[0044] In one embodiment, steps 301 to 305 are described as follows:

[0045] Step 301: Map the main search intent based on the domain knowledge graph to obtain a top-level node set, and extract the association relationship between the sub-search intent and the top-level node set based on the domain knowledge graph to obtain a node subset.

[0046] Optionally, after the financial data retrieval system determines the domain knowledge graph, since the nodes of the knowledge graph represent various financial concepts, the edges represent the relationship between concepts, and the main search intent represents the core direction of the user's search needs, the main search intent is mapped to the knowledge graph, which means finding the corresponding top-level concept nodes. These nodes are relatively broad and abstract concepts in the knowledge graph. For example, in the financial knowledge graph, "financial statement analysis" may correspond to top-level nodes such as "financial statements" and "financial indicator analysis", thus forming the top-level node set C. main Sub-search intents are refinements of the main search intent. The knowledge graph searches for associations between sub-search intents and nodes in the top-level node set. For example, "balance sheet" is directly related to "financial statements," and "payoff indicator" is related to "financial indicator analysis." This extracts nodes related to the sub-search intent, forming a node subset. These nodes reflect the sub-search intent's position and association within the knowledge graph.

[0047] Step 302: Determine the sub-intent weight of each sub-search intent based on the importance of the sub-search intent to the main search intent.

[0048] Optionally, since different sub-search intents play different roles in achieving the main search intent in the user's search needs, the financial data retrieval system can evaluate the contribution of each sub-search intent to the realization of the main search intent from multiple dimensions such as semantic relevance and business logic importance when analyzing the contribution of each sub-search intent to the realization of the main search intent. For example, if the main search intent is "to evaluate the profitability of the enterprise", the sub-search intent "to analyze the composition of the main business income" may be crucial for evaluating profitability, because the main business income is a key component of profitability, and the sub-search intent "to calculate the purchase cost of office supplies" is also related to the enterprise cost, but has less direct impact on profitability. Therefore, different weights can be assigned to different types of sub-search intents based on manually set financial expertise and experience. For example, the sub-intention involving the analysis of core financial indicators has a high weight, while the sub-intention involving only the viewing of basic report information has a low weight; the relationship between sub-search intent and the user's final satisfaction can also be learned through machine learning algorithms (such as multi-attribute decision analysis), and a weight w is assigned to each sub-search intent. i , the weight range is [0-1], the higher the priority, the greater the weight.

[0049] In one embodiment, taking the aforementioned primary search intent of "assessing financial status" as an example, the sub-search intents are "analyzing the company's 2024 debt-to-asset ratio" and "viewing the balance sheet page number." Based on financial professional experience, "analyzing the company's 2024 debt-to-asset ratio" is crucial for assessing financial status and is assigned a weight of 0.8; "viewing the balance sheet page number" is relatively unimportant and is assigned a weight of 0.2.

[0050] Step 303, starting from a node in the top-level node set, diffuses to the lower-level nodes along the edge of the domain knowledge graph to determine the semantic similarity between each term to be disambiguated and each node on the diffusion path, as well as the sub-intent level to which each node belongs and the corresponding sub-intent weight; wherein the term to be disambiguated is any term in the multi-layer retrieval intent.

[0051] Optionally, the financial data retrieval system obtains the top node set C according to step 301. main, starting from each node in the top-level node set, traverse and diffuse to the lower nodes along the edges between the nodes in the knowledge graph. During the diffusion process, for each term to be disambiguated in the multi-layer retrieval intent (for example, the word "control" in "cost control" may have different meanings in different contexts), calculate its semantic similarity with the concepts represented by each node on the diffusion path. You can use a path length-based method (such as the shorter the shortest path distance between two concepts in the knowledge graph, the higher the semantic similarity), or a vector embedding-based method (map the concept to the vector space, and measure the semantic similarity by calculating the cosine similarity between the vectors). At the same time, record the sub-intention level to which each node belongs (that is, which sub-retrieval intent the node is associated with) and the corresponding sub-intention weight.

[0052] Furthermore, let the term to be disambiguated be t and the top node set be C main , the node on the diffusion path is c i The semantic similarity calculation based on vector embedding can be calculated using the cosine similarity formula in and are the disambiguation term t and node c respectively i The vector representation of . At the same time, let the node c i The sub-intention level is l i , the corresponding sub-intention weight is

[0053] In one embodiment, for the term "liabilities" to be disambiguated under the primary search intent of "financial status assessment," the search begins with the top-level node "asset-liability analysis" and expands to include lower-level nodes such as "current liabilities" and "long-term liabilities." Using vector embedding, the semantic similarity between "liabilities" and "current liabilities" is calculated to be 0.9, and the semantic similarity between "liabilities" and "long-term liabilities" is 0.8. "Current liabilities" belongs to the sub-intent level of "Analyze the company's debt-to-asset ratio in 2024," with a corresponding sub-intent weight of 0.8; "long-term liabilities" also belongs to this sub-intent level and has a corresponding sub-intent weight of 0.8.

[0054] Step 304 , for the node set obtained by diffusion, determine the association strength between the node and the statement to be disambiguated and the hierarchical depth of the node in the domain knowledge graph, and filter the node set based on the association strength and hierarchical depth to obtain a candidate node set.

[0055] Optionally, the financial data retrieval system first determines the strength of association between each node and the statement to be disambiguated (i.e., the retrieval intent statement containing the term to be disambiguated) for the node set diffused in step 303. The strength of association between the node and the statement to be disambiguated can be determined based on the weight of the edge in the knowledge graph, and the weight of the edge usually reflects the closeness of the association between concepts. For example, in the financial knowledge graph, the weight of the edge between "asset-liability ratio" and "liability" may be high, indicating that they are closely associated. At the same time, the hierarchical depth of each node in the domain knowledge graph is determined. The deeper the hierarchical depth, the more specific and detailed the concept. The hierarchical depth represents the number of layers from the node to the top node. Nodes with too deep a hierarchy may contain concepts that are too detailed and not closely related to the user's main needs. Appropriate association strength thresholds and hierarchical depth thresholds are pre-set to filter out nodes with low association strength and too deep a hierarchy, and retain more relevant nodes as candidate node sets C. can .

[0056] Furthermore, let the node set obtained by diffusion be C = {c1, c2, ..., c n}, node c i The correlation strength with the sentence to be disambiguated is Rel(t,c i ) (obtained through the edge weight of the knowledge graph), node c i The depth of the hierarchy in the domain knowledge graph is Depth(c i ). Set the association strength threshold to τ rel , the layer depth threshold is τ depth , then the candidate node set C can ={c i ∣Rel(t,c i )≥τ rel ∧Depth(c i )≤τ deph ,c i ∈C}.

[0057] In one embodiment, for the node set related to "Liabilities" obtained through diffusion, the association strength (based on edge weights) between "Accounts Payable" and "Liabilities" is 0.7, and the association strength between "Employee Salaries and Bonuses Payable" and "Liabilities" is 0.3. The hierarchical depth of "Accounts Payable" is 2, and the hierarchical depth of "Employee Salaries and Bonuses Payable" is 3. Assuming the association strength threshold is set to 0.4 and the hierarchical depth threshold is set to 3, "Employee Salaries and Bonuses Payable" is filtered out, while nodes that meet the criteria, such as "Accounts Payable," are retained in the candidate node set.

[0058] Step 305 : For each node in the candidate node set, a comprehensive score is determined based on a preset comprehensive score function, and the candidate node set is screened based on the comprehensive score to obtain a term disambiguation result.

[0059] Optionally, the financial data retrieval system calculates the semantic similarity Sim(t,c i ), sub-intent weight Relation strength Rel(t,c i ) and other information, and substitute them into the preset comprehensive score function: Calculate the comprehensive score, where c represents any node in the candidate node set; n represents the number of sub-meaning hierarchical graphs associated with candidate node c; w i Expressed as the sub-intention weight of the i-th sub-intention level; Sim(t,c i ) represents the term to be disambiguated t and the corresponding node c of the i-th sub-intention level i Semantic similarity of Depth(c i ) represents the corresponding node c of the i-th sub-intention level i Hierarchical depth in the domain knowledge graph; Rel(t,c i ) represents the corresponding node c of the i-th sub-intention level i The strength of the association with the term to be disambiguated t. A higher score indicates a closer match between the concept represented by the node and the term to be disambiguated. The candidate nodes are then sorted based on their combined scores, and the concept represented by the node with the highest score is selected as the term disambiguation result. If the combined scores are all below a preset threshold, manual verification of multiple candidates may be triggered, with further human judgment on the accuracy of the disambiguation result.

[0060] In one embodiment, there are "inventory" and "accounts receivable" nodes in the candidate node set. For the "inventory" node, the function The calculated overall score is 0.6; for the "Accounts Receivable" node, the calculated overall score is 0.3. The "Inventory" node has a higher score, so "Inventory" is used as the disambiguation result for the "Inventory" term in "Inventory-Related Financial Indicators."

[0061] The embodiment of the present invention maps the top-level node set from the main search intent, and then extracts the node subset in combination with the sub-search intent association relationship, which can quickly locate the area related to the search intent in the knowledge graph, provide an accurate search range for term disambiguation, and improve the disambiguation efficiency; and consider multiple dimensional factors such as sub-intent weight, semantic similarity, association strength and hierarchical depth, comprehensively evaluate the meaning of the term, avoid disambiguation deviation caused by a single factor, and make the disambiguation result more accurate and reliable; in addition, through the association strength and hierarchical depth to filter out noise concepts, reduce the interference of irrelevant information, and further improve the quality and efficiency of disambiguation, finally screen the candidate nodes based on the comprehensive score, and determine the disambiguation result in a quantitative manner to ensure the objectivity of the result. At the same time, the manual verification mechanism provides a reliable solution for special cases.

[0062] In one embodiment, steps 401 to 403 are described as follows:

[0063] Step 401: All semantic units in the term disambiguation result are used as candidate nodes of a logic tree. For each pair of candidate nodes, a connectivity index is determined based on a connectivity function.

[0064] Optionally, the financial data retrieval system first extracts all semantic units from the term disambiguation results as candidate nodes for the logic tree. For example, if the term disambiguation result is "Analyze the cost impact of straight-line depreciation of fixed assets in an enterprise," the semantic units "enterprise," "fixed assets," "straight-line depreciation," and "cost impact" will all become candidate nodes. Then, for each pair of candidate nodes, according to the connectivity function: Calculate the connectivity index. Among them, the shortest path length between nodes n1 and n2 is determined by the graph structure algorithm of the knowledge graph (such as Dijkstra algorithm); the semantic similarity calculation method (such as cosine similarity based on word vectors) is used to determine the semantic similarity S lian (n1, n2); According to the user's search intention analysis, determine the intention dependency D of nodes n1 and n2 lian For example, "fixed assets" and "straight-line depreciation" are connected by related paths in the knowledge graph. The shortest path length between them is calculated, and the semantic similarity between the two concepts is calculated using word vectors. The intent dependency is then determined based on the close association between the two concepts in the user's search intent, ultimately yielding their connectivity index.

[0065] In one embodiment, assume that the term disambiguation result includes three semantic units: "net profit," "operating income," and "cost," as candidate nodes. For the candidate node pair "net profit" and "operating income," the shortest path length between them in the financial knowledge graph is calculated to be 3 (assuming the path length between nodes in the graph is used as the measure), the semantic similarity calculated using vectors is 0.6, and the intent dependency is determined to be 0.7 based on business logic. According to the connectivity function: Available connectivity index

[0066] Step 402: Combine candidate nodes from high to low based on connectivity index to obtain a tree topology structure.

[0067] Optionally, the financial data retrieval system combines candidate nodes in descending order based on the connectivity index calculated in step 401. A high connectivity index means that the nodes are semantically closely related and highly interdependent in satisfying user intent. Nodes with high connectivity are combined first, gradually constructing a tree-like topology. For example, the "net profit" and "cost" nodes with the highest connectivity are combined first, as they are closely related in financial analysis. Other nodes are then added in sequence to form a tree-like structure with clear hierarchies and close node connections.

[0068] In one embodiment, continuing with the example of step 401, assume that the connectivity index between "net profit" and "cost" is the highest, at 5.5; the connectivity index between "net profit" and "operating income" is 7.14; and the connectivity index between "operating income" and "cost" is 8.2. First, combine "net profit" and "cost," then add "operating income" in an appropriate manner to form a tree topology, with "net profit" as the parent node and "cost" and "operating income" as child nodes (the specific structure is determined by business logic and connectivity relationships).

[0069] Step 403 : simplify and delete the tree topology structure, and construct a query retrieval logic tree with the candidate node having the highest connectivity index as the root node.

[0070] Optionally, after the financial data retrieval system completes the tree topology, some edges or nodes may still have low connectivity and contribute little to the overall structure and the expression of user intent. Therefore, a connectivity threshold is set to remove edges with connectivity indicators below the threshold, simplifying the tree structure and reducing overcomplexity. At the same time, the node with the highest connectivity index and consistent with the main graph is selected from the candidate nodes as the root node. This is because the root node is the core of the logical tree. High connectivity and consistency with the main graph can better guide the entire logical tree, allowing it to develop around the core concept and more accurately express the user's search intent.

[0071] In one embodiment, the connectivity threshold is set to 7. In the above example, the connectivity index between "operating income" and "cost" is 8.2, which is above the threshold and retained; the connectivity index between "net profit" and "operating income" is 7.14, which is above the threshold and retained; the connectivity index between "net profit" and "cost" is 5.5, which is below the threshold and deleted. Assuming that the connectivity index of "net profit" is the highest among all nodes and matches the main graph "Financial Profit Analysis", "net profit" is used as the root node to construct the final query retrieval logic tree, with "operating income" and "cost" as child nodes under the "net profit" node.

[0072] The embodiment of the present invention starts from the semantic units after disambiguation and constructs a logical tree based on the connectivity function, which can accurately map the semantics and concept associations in the user's search intention, so that the logical tree accurately expresses the user's needs; and through the connection index sorting and combining nodes and simplifying the deletion operation, a logical tree with a reasonable structure, concise and clear structure is constructed, which not only highlights key concepts but also avoids redundancy, thereby improving retrieval efficiency and accuracy.

[0073] In one embodiment, steps 501 to 504 are described as follows:

[0074] Step 501 , converting the main search intent, sub-search intent and preliminary search results into semantic vectors respectively to obtain the main search intent semantic vector, the sub-search intent semantic vector and the preliminary search result semantic vector.

[0075] Optionally, the financial data retrieval system utilizes word vector models such as Word2Vec or GloVe to convert the keywords in the main search intent, sub-search intents, and text information in the preliminary search results into semantic vectors. For example, for the main search intent "analyze the financial status of an enterprise," keywords such as "enterprise" and "financial status" can be converted into semantic vectors; for the sub-search intent "calculate profit margin," "profit margin" can be converted into a semantic vector; and if the preliminary search results contain "enterprise net profit margin data for 2024," the relevant keywords can also be converted into semantic vectors. In this way, text information is converted into a numerical vector form that can be processed and compared by computers, paving the way for subsequent similarity calculations.

[0076] In one embodiment, assuming the main search intent is "analyzing the company's revenue situation", the Word2Vec model is used to convert the company's "and revenue situation" into a semantic vector. and The sub-search intent is to view quarterly revenue data, and convert "quarterly" and "revenue data" into semantic vectors. and The initial search result is the company's second quarter 2024 revenue report. Convert "company", "2024", "second quarter", and "revenue report" into semantic vectors.

[0077] In step 502, similarity processing is performed on the main search intent semantic vector and the sub-search intent semantic vector respectively with the preliminary search result semantic vector to obtain a first similarity score and a second similarity score.

[0078] Optionally, the financial data retrieval system usually adopts the cosine similarity calculation method when calculating the similarity, that is, the similarity between two vectors is measured by calculating the cosine value of the angle between them. The closer the value is to 1, the more similar the two vectors are. Specifically, for the main search intent semantic vector and the preliminary search result semantic vector, the cosine similarity between them is calculated to obtain a first similarity score. This score reflects the degree of matching between the preliminary search result and the main search intent. Similarly, for each sub-search intent semantic vector and the preliminary search result semantic vector, the cosine similarity is calculated separately to obtain multiple second similarity scores. These scores reflect the degree of matching between the preliminary search result and each sub-search intent.

[0079] In one embodiment, embodiment: Continuing with the above example, the main search intent semantic vector (Depend on and etc.) and the semantic vector of the preliminary retrieval results (Depend on etc.), according to the cosine similarity formula Calculate the first similarity score, assuming it is 0.7. Sub-retrieval intent semantic vector and The second similarity score is calculated to be 0.6. and The second similarity score is calculated to be 0.5.

[0080] Step 503: construct a similarity matrix based on the first similarity score and the second similarity score; each row in the similarity matrix represents a preliminary search result, and each column corresponds to the similarity score of the main search intent and each sub-search intent.

[0081] Optionally, after obtaining the similarity scores between all preliminary search results and the main search intent and sub-search intent, the financial data retrieval system organizes these scores into a matrix. Each row of the matrix corresponds to a preliminary search result, and each column corresponds to the similarity score of the main search intent or a sub-search intent. For example, assuming there are m preliminary search results and the main search intent is m, the financial data retrieval system organizes these scores into a matrix. Figure 1 If there are k sub-search intents, the constructed similarity matrix is ​​an m*(k+1) matrix. This can clearly show the similarity between each preliminary search result and different search intents, facilitating subsequent sorting operations.

[0082] In one embodiment, it is assumed that there are three preliminary search results R1, R2, and R3, and the main search intent is I main , the sub-search intent is I sub1 , I sub2 . R1 and I main The first similarity score is 0.7, which is the same as I sub1The second similarity score is 0.6, which is the same as I sub2 The second similarity score is 0.5; R2 and I main The first similarity score is 0.6, which is the same as I sub1 The second similarity score is 0.4, which is the same as I sub2 The second similarity score of R3 is 0.3; main The first similarity score is 0.8, which is the same as I sub1 The second similarity score is 0.7, which is the same as I sub2 The second similarity score is 0.6. The similarity matrix constructed is as follows:

[0083] Main search intent similarity <![CDATA[I sub1 Similarity]]> <![CDATA[I sub2 Similarity]]> <![CDATA[R1]]> 0.7 0.6 0.5 <![CDATA[R2]]> 0.6 0.4 0.3 <![CDATA[R3]]> 0.8 0.7 0.6

[0084] Step 504 : weighting and sorting the preliminary search results in the similarity matrix based on a preset nested sorting strategy to obtain a fused search result.

[0085] Optionally, the nested sorting strategy preset by the financial data retrieval system is formulated based on the importance and hierarchical relationship of the main search intent and the sub-search intent. Generally speaking, the first similarity score (i.e., the similarity score with the main search intent) is first sorted in descending order. This is because the main search intent represents the main direction of the user's search, and the preliminary search results with a high degree of match with the main search intent should be prioritized. When the first similarity scores are the same, further sorting is performed according to the second similarity score (the similarity score with the sub-search intent). A weighted sum or sequential comparison method can be set according to the priority order of the sub-search intent. For example, first compare the second similarity score corresponding to the important sub-search intent. If they are still the same, compare the scores of other sub-search intents, and finally obtain a fused search result so that the search results are arranged from high to low according to the degree of match with the user's search intent.

[0086] In one embodiment, for the above similarity matrix, according to the nested sorting strategy, it is first sorted in descending order according to the main search intent similarity score, with R3 (0.8) at the top, followed by R1 (0.7), and finally R2 (0.6). In this example, the first similarity score is already able to distinguish the order, and there is no need to further compare based on the sub-search intent similarity score. If there is a situation where the first similarity score is the same, such as R4, whose main search intent similarity score is also 0.7, then it is necessary to further compare based on the sub-search intent similarity score. Assuming that the sub-search intent I sub1 Priority higher than I sub2 , R1 and I sub1 The similarity score of 0.6 is higher than that of R4 and I sub1If the similarity score is 0.5, R1 is ranked before R4, and the final fusion search results are R3, R1, and R2 (assuming R4 does not exist in the final result).

[0087] The embodiment of the present invention deeply explores the semantic association between retrieval intent and retrieval results through semantic vector conversion and similarity calculation, avoids the limitation of sorting only from surface keyword matching, and improves the understanding and processing ability of semantics; and based on the hierarchical structure of main retrieval intent and sub-retrieval intent, adopts a nested sorting strategy to gradually and accurately sort the retrieval results from the whole to the part, so that the sorting results are more in line with the logic of user retrieval needs, and improves the quality and practicality of the retrieval results; finally, the similarity information is structured through the construction of a similarity matrix, which provides a clear and orderly data structure for the sorting operation, facilitates numerical calculation and comparison, and improves the efficiency and accuracy of sorting.

[0088] Furthermore, the artificial intelligence-based financial data retrieval system provided by the present invention is described below. The artificial intelligence-based financial data retrieval system described below and the artificial intelligence-based financial data retrieval method described above can refer to each other.

[0089] Optional, see Figure 2 , Figure 2 : is a schematic diagram of the structure of the artificial intelligence-based financial data retrieval system provided by the present invention. The artificial intelligence-based financial data retrieval system includes:

[0090] The user profile building module 210 is used to obtain the original query data and associated information data input by the user; and input the associated information data into the user profile model to obtain the user category output by the user profile model; the user profile model is trained based on the sample associated information and its corresponding user category label results;

[0091] The intent recognition module 220 is used to input the user category and the original query data into the intent recognition model to obtain the multi-layered retrieval intent output by the intent recognition model; the multi-layered retrieval intent includes the main retrieval intent and the sub-retrieval intent; the intent recognition model is trained based on the sample user category, the sample query data and the corresponding retrieval intent label results;

[0092] A term disambiguation module 230 is configured to determine a domain knowledge graph based on user categories, and to disambiguate terms in the multi-layer search intent based on the domain knowledge graph to obtain a term disambiguation result;

[0093] A reconstructing retrieval module 240 is configured to reconstruct a query retrieval logic tree based on the term disambiguation result, and determine preliminary retrieval results based on the query retrieval logic tree;

[0094] The search screening module 250 is used to fuse and sort the preliminary search results based on the multi-layer search intent to obtain a fused search result, and to screen the fused search result based on the user category to determine the target search result.

[0095] The embodiment of the present invention constructs a user profile by acquiring related information data, and can use different knowledge graphs for term disambiguation for different user categories, effectively solving the semantic understanding problems caused by the ambiguity and specificity of financial terms and the diversity of user expressions, and improving the accuracy of natural language processing in financial data retrieval; in addition, through the intention recognition model, multi-layer retrieval intentions are derived, and the retrieval results are integrated and sorted based on this, which can accurately distinguish the different intentions behind similar retrieval content of different users, avoid repeated retrieval, improve retrieval efficiency, and reduce the waste of computing resources; finally, from user portrait construction, intention recognition, term disambiguation to retrieval result processing, and combined with user category for personalized processing, it ensures that different types of users can obtain target retrieval results that meet their own needs, thereby improving the speed and quality of users' acquisition of effective information.

[0096] See also Figure 3 , Figure 3 This is a diagram of an embodiment of an electronic device provided by an embodiment of the present invention. Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, the following steps are implemented:

[0097] Obtain the original query data and associated information data input by the user; input the associated information data into the user portrait model to obtain the user category output by the user portrait model; the user portrait model is trained based on the sample associated information and its corresponding user category label results;

[0098] Input user categories and original query data into the intent recognition model to obtain the multi-layer retrieval intent output by the intent recognition model; the multi-layer retrieval intent includes the main retrieval intent and sub-retrieval intent; the intent recognition model is trained based on sample user categories, sample query data and their corresponding retrieval intent label results;

[0099] Based on the user category, the domain knowledge graph is determined, and the terms in the multi-layer search intent are disambiguated based on the domain knowledge graph to obtain the term disambiguation results;

[0100] Reconstructing the query retrieval logic tree based on the term disambiguation results, and determining preliminary retrieval results based on the query retrieval logic tree;

[0101] The preliminary search results are fused and sorted based on multi-layer search intent to obtain fused search results, and the fused search results are filtered based on user categories to determine the target search results.

[0102] See also Figure 4 , Figure 4 Detailed description of an embodiment of a computer-readable storage medium provided by an embodiment of the present invention. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, the following steps are implemented:

[0103] Obtain the original query data and associated information data input by the user; input the associated information data into the user portrait model to obtain the user category output by the user portrait model; the user portrait model is trained based on the sample associated information and its corresponding user category label results;

[0104] Input user categories and original query data into the intent recognition model to obtain the multi-layer retrieval intent output by the intent recognition model; the multi-layer retrieval intent includes the main retrieval intent and sub-retrieval intent; the intent recognition model is trained based on sample user categories, sample query data and their corresponding retrieval intent label results;

[0105] Based on the user category, the domain knowledge graph is determined, and the terms in the multi-layer search intent are disambiguated based on the domain knowledge graph to obtain the term disambiguation results;

[0106] Reconstructing the query retrieval logic tree based on the term disambiguation results, and determining preliminary retrieval results based on the query retrieval logic tree;

[0107] The preliminary search results are fused and sorted based on multi-layer search intent to obtain fused search results, and the fused search results are filtered based on user categories to determine the target search results.

[0108] In another aspect, the present invention further provides a computer program product, which includes a computer program. The computer program may be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the artificial intelligence-based financial data retrieval method provided by the above methods, which includes:

[0109] Obtain the original query data and associated information data input by the user; input the associated information data into the user portrait model to obtain the user category output by the user portrait model; the user portrait model is trained based on the sample associated information and its corresponding user category label results;

[0110] Input user categories and original query data into the intent recognition model to obtain the multi-layer retrieval intent output by the intent recognition model; the multi-layer retrieval intent includes the main retrieval intent and sub-retrieval intent; the intent recognition model is trained based on sample user categories, sample query data and their corresponding retrieval intent label results;

[0111] Based on the user category, the domain knowledge graph is determined, and the terms in the multi-layer search intent are disambiguated based on the domain knowledge graph to obtain the term disambiguation results;

[0112] Reconstructing the query retrieval logic tree based on the term disambiguation results, and determining preliminary retrieval results based on the query retrieval logic tree;

[0113] The preliminary search results are fused and sorted based on multi-layer search intent to obtain fused search results, and the fused search results are filtered based on user categories to determine the target search results.

[0114] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units. That is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0115] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or certain parts of the embodiment.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A financial data retrieval method based on artificial intelligence, characterized in that: include: Obtain the original query data and related information data entered by the user; and inputting the associated information data into a user portrait model to obtain a user category output by the user portrait model; The user portrait model is trained based on sample association information and its corresponding user category label results; Inputting the user category and the original query data into an intent recognition model to obtain a multi-layer retrieval intent output by the intent recognition model; The multi-layered search intent includes a main search intent and a sub-search intent; The intent recognition model is trained based on sample user categories, sample query data and their corresponding retrieval intent label results; Based on the user category, a domain knowledge graph is determined, and terms in the multi-layer search intent are disambiguated based on the domain knowledge graph to obtain a term disambiguation result; reconstructing a query retrieval logic tree based on the term disambiguation result, and determining a preliminary retrieval result based on the query retrieval logic tree; The preliminary search results are fused and sorted based on the multi-layer search intent to obtain a fused search result, and the fused search result is screened based on the user category to determine the target search result.

2. The financial data retrieval method based on artificial intelligence according to claim 1, characterized in that: The associated information data includes user identity tags, historical search records, and current operation scenarios, wherein the user identity tags include user IDs and their corresponding identity permissions.

3. The financial data retrieval method based on artificial intelligence according to claim 1, characterized in that: The disambiguating the terms in the multi-layer search intent based on the domain knowledge graph to obtain a term disambiguation result includes: Mapping the main search intent based on the domain knowledge graph to obtain a top-level node set, and extracting the association relationship between the sub-search intent and the top-level node set based on the domain knowledge graph to obtain a node subset; Determining a sub-intention weight of each sub-search intent based on the importance of the sub-search intent to the main search intent; Starting from a node in the top-level node set, diffusion is performed along the edge of the domain knowledge graph to the lower-level nodes, and the semantic similarity between each term to be disambiguated and each node on the diffusion path, as well as the sub-intent level to which each node belongs and the corresponding sub-intent weight are determined; the term to be disambiguated is any term in the multi-layer retrieval intent; For the node set obtained by diffusion, determine the association strength between the node and the statement to be disambiguated and the hierarchical depth of the node in the domain knowledge graph, and filter the node set based on the association strength and hierarchical depth to obtain the candidate node set; For each node in the candidate node set, a comprehensive score is determined based on a preset comprehensive score function, and the candidate node set is screened based on the comprehensive score to obtain the term disambiguation result.

4. The financial data retrieval method based on artificial intelligence according to claim 3, characterized in that: The comprehensive function is: Where c represents any node in the candidate node set; n represents the number of sub-meaning hierarchical graphs associated with candidate node c; w i Expressed as the sub-intention weight of the i-th sub-intention level; Sim(t,c i ) represents the term to be disambiguated t and the corresponding node c of the i-th sub-intention level i Semantic similarity of Depth(c i ) represents the corresponding node c of the i-th sub-intention level i Hierarchical depth in the domain knowledge graph; Rel(t,c i ) represents the corresponding node c of the i-th sub-intention level i The strength of the association with the term to be disambiguated t.

5. The financial data retrieval method based on artificial intelligence according to claim 1, characterized in that: The step of reconstructing the query retrieval logic tree based on the term disambiguation result includes: Taking all semantic units in the term disambiguation result as candidate nodes of a logic tree, and determining a connectivity index for each pair of candidate nodes based on a connectivity function; Combining candidate nodes from high to low based on the connectivity index to obtain a tree topology structure; The tree topology structure is simplified and deleted, and the query retrieval logic tree is constructed by taking the candidate node with the highest connectivity index as the root node.

6. The financial data retrieval method based on artificial intelligence according to claim 5, characterized in that: The connectivity function is: Among them, C lian (n1,n2) represents the connectivity index of nodes n1 and n2; L lian (n1,n2) represents the shortest path length between nodes n1 and n2; S lian (n1, n2) represents the semantic similarity between nodes n1 and n2; D lian (n1,n2) is represented as the intention dependency graph of nodes n1 and n2.

7. The financial data retrieval method based on artificial intelligence according to claim 1, characterized in that: The preliminary search results are fused and sorted based on the multi-layer search intent to obtain a fused search result, including: Convert the main search intent, sub-search intent and preliminary search results into semantic vectors respectively to obtain the main search intent semantic vector, the sub-search intent semantic vector and the preliminary search result semantic vector; Performing similarity processing on the main search intent semantic vector and the sub-search intent semantic vector with the preliminary search result semantic vector respectively to obtain a first similarity score and a second similarity score; Based on the first similarity score and the second similarity score, construct a similarity matrix; each row in the similarity matrix represents a preliminary search result, and each column corresponds to the similarity score of the main search intent and each sub-search intent; The preliminary search results in the similarity matrix are weighted and sorted based on a preset nested sorting strategy to obtain the fused search results.

8. A financial data retrieval system based on artificial intelligence, characterized in that: Applicable to the artificial intelligence-based financial data retrieval method according to any one of claims 1 to 7; the artificial intelligence-based financial data retrieval system comprises: A user profile building module is used to obtain the original query data and associated information data input by the user; and input the associated information data into the user profile model to obtain the user category output by the user profile model; the user profile model is trained based on the sample associated information and its corresponding user category label results; An intent recognition module is configured to input the user category and the original query data into an intent recognition model to obtain a multi-layered retrieval intent output by the intent recognition model; the multi-layered retrieval intent includes a main retrieval intent and sub-retrieval intents; the intent recognition model is trained based on sample user categories, sample query data, and their corresponding retrieval intent label results; A term disambiguation module is used to determine a domain knowledge graph based on user categories, and to disambiguate terms in the multi-layer search intent based on the domain knowledge graph to obtain a term disambiguation result; a reconstructing retrieval module, configured to reconstruct a query retrieval logic tree based on the term disambiguation result, and determine a preliminary retrieval result based on the query retrieval logic tree; The retrieval screening module is used to fuse and sort the preliminary retrieval results based on multi-layer retrieval intentions to obtain fused retrieval results, and to screen the fused retrieval results based on user categories to determine target retrieval results.

9. An electronic device comprising: Memory for storing computer software programs; A processor for reading and executing the computer software program, wherein when the processor executes the computer software program, it implements the artificial intelligence-based financial data retrieval method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing a computer software program, wherein: When the computer software program is executed by a processor, it implements the artificial intelligence-based financial data retrieval method as described in any one of claims 1 to 7.

Images