Meta analysis method, system and equipment based on artificial intelligence and storage medium

Abstract

The invention discloses a Meta analysis method, system and device based on artificial intelligence and a storage medium, and the method comprises the steps: receiving a natural language research problem inputted by a user, analyzing the natural language research problem, and generating a structured research framework according to an analysis result; constructing a retrieval formula based on the structured research framework, and retrieving literatures from the plurality of literature databases by using the retrieval formula to obtain a candidate literature set; and performing information extraction on the candidate literature set to obtain quantitative research data, and performing Meta analysis based on the quantitative research data to obtain a Meta analysis result. According to the method, the natural language research problem is analyzed, the structured research framework is generated according to the analysis result, the retrieval formula is constructed based on the structured research framework to retrieve literatures from the multiple literature databases, information extraction is performed on the candidate literature set, Meta analysis is performed based on the extracted quantitative research data, and the efficiency and accuracy of Meta analysis are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a meta-analysis method, system, device and storage medium based on artificial intelligence. Background Technology

[0002] Meta-analysis, as an important statistical method, aims to integrate the results of multiple independent studies in a systematic and quantitative manner, thereby drawing more universal and persuasive conclusions. It has been widely applied in many fields such as evidence-based medicine, drug efficacy evaluation, and epidemiological surveys, providing crucial scientific evidence for clinical decision-making and health policy formulation. However, traditional meta-analysis is cumbersome, time-consuming, and error-prone, involving multiple complex steps such as literature screening, data extraction, quality assessment, and statistical analysis.

[0003] First, in the literature retrieval and screening stage, the construction of search queries heavily relies on the researchers' expertise and requires adjusting syntax across different databases (such as PubMed and Embase), a tedious process that can easily lead to overlooking key studies or introducing irrelevant literature. Subsequently, screening a massive amount of literature is an extremely time-consuming and labor-intensive task, and the subjectivity of manual judgment can introduce selection bias. Second, in the data extraction stage, key data such as sample size, mean, standard deviation, and effect size need to be manually extracted from tables, charts, or text in full-text articles (usually in PDF format). This process is not only prone to errors due to fatigue, but also, when the original research data is not presented in a standardized manner, different extractors may obtain different results, seriously affecting the accuracy of the data.

[0004] Therefore, there is an urgent need for an AI-based meta-analysis method that can improve the efficiency and accuracy of meta-analysis. Summary of the Invention

[0005] The main objective of this invention is to provide an artificial intelligence-based meta-analysis method, system, device, and storage medium, aiming to solve the technical problems of low efficiency and insufficient accuracy of traditional meta-analysis methods.

[0006] To achieve the above objectives, this invention provides an artificial intelligence-based meta-analysis method, which includes the following steps: The system receives natural language research questions input by users, parses the natural language research questions, and generates a structured research framework based on the parsing results. Based on the structured research framework, a search formula is constructed, and the search formula is used to retrieve literature from multiple literature databases to obtain a candidate literature set; Information is extracted from the candidate literature set to obtain quantitative research data, and meta-analysis is performed based on the quantitative research data to obtain the meta-analysis results.

[0007] Optionally, the step of extracting information from the candidate literature set to obtain quantitative research data, and performing meta-analysis based on the quantitative research data to obtain meta-analysis results includes: Extract the title and abstract information of each document in the candidate document set; Based on the title and abstract information, a large language model is used to determine whether the corresponding documents in the candidate document set meet the preset inclusion criteria. The literature that meets the preset inclusion criteria in the candidate literature set is taken as the target literature set; Information is extracted from the target literature set to obtain quantitative research data, and meta-analysis is performed based on the quantitative research data to obtain the meta-analysis results.

[0008] Optionally, after the step of selecting the candidate document set that meets the preset inclusion criteria as the target document set, the method further includes: Obtain the text data of each document in the target document set; Based on the text data, the large language model is used to determine whether the corresponding documents in the target document set meet the preset inclusion criteria. The literature that meets the preset inclusion criteria in the target literature set is used as the filtered target literature set.

[0009] Optionally, after the step of retrieving documents from multiple document databases using the search query to obtain a candidate document set, the method further includes: The candidate document set is deduplicated to obtain a deduplicated candidate document set, and the title and abstract information of each document in the deduplicated candidate document set are extracted. The title and abstract information of each document in the deduplicated candidate document set are used as the information to be evaluated; The information to be evaluated is converted into a corresponding document vector, and the structured research framework is converted into a structured research framework vector. The similarity between each document vector and the structured research framework vector is determined, and the feasibility of the deduplicated candidate document set is evaluated based on the similarity to obtain the evaluated candidate document set.

[0010] Optionally, the step of constructing a search query based on the structured research framework, and using the search query to retrieve literature from multiple literature databases to obtain a candidate literature set includes: The elements in the structured research framework are expanded into corresponding topic words and free words using a large language model; A search query is generated based on the logical combination of the subject terms and free terms, and the search query is converted into a query request information in a standard query format. The query request information is sent to multiple literature databases through the application programming interface to obtain a set of candidate documents.

[0011] Optionally, the step of performing meta-analysis based on the quantitative research data to obtain meta-analysis results includes: Based on the quantitative research data, multiple effect sizes are determined, and heterogeneity tests are performed on each effect size to obtain the test results; Based on the test results, a corresponding analytical model is selected, and the analytical model is used to combine multiple effect sizes to obtain the combined effect size. Publication bias detection and subgroup analysis were performed based on the various effect sizes to obtain publication bias detection results and subgroup analysis results; The test results, the pooled effect size, the publication bias detection results, and the subgroup analysis results were used as the results of the meta-analysis.

[0012] Optionally, the step of receiving a natural language research question input by a user, parsing the natural language research question, and generating a structured research framework based on the parsing result includes: Receive a natural language research question input by the user and obtain the context information corresponding to the natural language research question; Based on the context information, a multi-turn conversational parsing is performed on the natural language research question to obtain the parsing results; Based on the parsing results, key information about the natural language research problem is determined, and a structured research framework is generated based on the key information.

[0013] Furthermore, to achieve the above objectives, this invention also proposes an artificial intelligence-based meta-analysis system, the system comprising: The problem parsing module is used to receive natural language research questions input by users, parse the natural language research questions, and generate a structured research framework based on the parsing results. The literature screening module is used to construct a search query based on the structured research framework, and use the search query to retrieve literature from multiple literature databases to obtain a candidate literature set; The information extraction module is used to extract information from the candidate literature set, obtain quantitative research data, and perform meta-analysis based on the quantitative research data to obtain meta-analysis results.

[0014] Furthermore, to achieve the above objectives, the present invention also proposes an artificial intelligence-based meta-analysis device, the device comprising: a memory, a processor, and an artificial intelligence-based meta-analysis program stored in the memory and executable on the processor, the artificial intelligence-based meta-analysis program being configured to implement the steps of the artificial intelligence-based meta-analysis method described above.

[0015] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing an artificial intelligence-based meta-analysis program, which, when executed by a processor, implements the steps of the artificial intelligence-based meta-analysis method described above.

[0016] This invention discloses a method for receiving natural language research questions input by users, parsing the natural language research questions, generating a structured research framework based on the parsing results, constructing a search query based on the structured research framework, retrieving literature from multiple literature databases using the search query to obtain a candidate literature set, extracting information from the candidate literature set to obtain quantitative research data, and performing meta-analysis based on the quantitative research data to obtain meta-analysis results. Because this invention parses natural language research questions, generates a structured research framework based on the parsing results, then constructs a search query based on the structured research framework to retrieve literature from multiple literature databases, and finally extracts information from the candidate literature set and performs meta-analysis based on the extracted quantitative research data, compared with existing technologies, this invention achieves end-to-end automation, improving the efficiency and accuracy of meta-analysis. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the AI-based meta-analysis method of the present invention. Figure 2 This is a flowchart illustrating the second embodiment of the AI-based meta-analysis method of the present invention. Figure 3 This is a flowchart illustrating the third embodiment of the AI-based meta-analysis method of the present invention. Figure 4 This is a structural block diagram of the first embodiment of the AI-based meta-analysis system of the present invention; Figure 5 This is a schematic diagram of the structure of an AI-based meta-analysis device in the hardware operating environment of the embodiment of the present invention.

[0018] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0020] This invention provides an artificial intelligence-based meta-analysis method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the meta-analysis method based on artificial intelligence of the present invention.

[0021] In this embodiment, the AI-based meta-analysis method includes steps S10 to S30: Step S10: Receive the natural language research question input by the user, parse the natural language research question, and generate a structured research framework based on the parsing result.

[0022] It should be noted that the executing entity in this embodiment can be a computer server device with data processing, network communication, and program execution functions applied in a meta-analysis scenario, such as a server, tablet computer, or personal computer, or an electronic device capable of performing the above functions (such as an AI-based meta-analysis device). The following uses an AI-based meta-analysis system as an example to illustrate this embodiment and the subsequent embodiments.

[0023] Understandably, natural language research questions can refer to scientific questions posed by users using everyday, unstructured language, which they hope to answer through meta-analysis. Unlike traditional systems that require users to fill in specific fields, this embodiment receives continuous, context-dependent text strings whose complete meaning depends on the inherent logic of natural language, rather than a pre-defined form structure. This effectively lowers the barrier to entry for users, who do not need to be familiar with professional meta-analysis formats (such as PICOS) beforehand and can directly describe their research intentions.

[0024] It should be noted that the parsing results can refer to the internal representation of key semantic elements and their logical relationships extracted after in-depth understanding and analysis of natural language research problems through large language models. The structured research framework can refer to a machine-readable data structure generated based on the parsing results, according to meta-analysis standards (such as the PICOS principle), and capable of driving subsequent automated processes.

[0025] It should be added that a structured research framework can refer to a structured data object, such as a JSON (JavaScript Object Notation) or XML file. Its content explicitly corresponds to the various dimensions of PICOS: P (Population): clearly defining the research subjects, such as patients or populations with a specific disease; I (Intervention): describing the intervention (such as drug treatment, surgery, etc.); C (Comparison): the control measure (such as existing therapies or placebo); O (Outcome): the study outcome indicators (such as efficacy, adverse reactions, etc.); S (Study design): the type of study design (such as randomized controlled trials, cohort studies, etc.).

[0026] In a specific implementation, a natural language research question input by a user can be received, and context information corresponding to the natural language research question can be obtained; based on the context information, the natural language research question can be parsed in a multi-turn conversational manner to obtain the parsing result; based on the parsing result, the key information of the natural language research question can be determined, and a structured research framework can be generated based on the key information.

[0027] It should be understood that contextual information refers to various relevant information, other than the question text itself, used to assist and refine the understanding of the user's intent during the process of parsing the user's natural language research questions. Contextual information can include multi-turn dialogue history and real-time feedback from the user to the system's questions. By introducing contextual information, the system's understanding is no longer based on a single sentence in isolation, but rather within a rich interactive environment, thereby generating more accurate parsing results that better meet the user's actual needs.

[0028] It should be noted that multi-turn conversational parsing refers to a method where the system engages in continuous dialogue with the user, gradually guiding the user to supplement information and clarify ambiguities through multiple question-and-answer iterations, thereby accurately completing the structured parsing of the research question. This effectively addresses the ambiguity and incompleteness of input in natural language research questions, ensuring through proactive interaction that the final generated structured research framework is accurate, complete, and operable, laying the foundation for subsequent full-process automation.

[0029] Understandably, by using contextual information to perform multi-turn conversational parsing of natural language research questions, the system can dynamically complete the missing PICOS elements in the user's single-turn input, significantly reducing the probability of erroneous parsing due to incomplete information, thereby improving the completeness and accuracy of the subsequent structured research framework.

[0030] It should be understood that key information can refer to the core structured data that is finally extracted from user input after multiple rounds of conversational parsing, conforms to the Meta-analysis specification (PICOS principle), and is used to drive all subsequent automation modules. Key information directly corresponds to the five dimensions of PICOS.

[0031] Step S20: Construct a search query based on the structured research framework, and use the search query to retrieve literature from multiple literature databases to obtain a candidate literature set.

[0032] It should be noted that the elements in the structured research framework can be expanded into corresponding subject terms and free terms using a large language model; a search query can be generated based on the logical combination of the subject terms and free terms, and the search query can be converted into a query request information in a standard query format; the query request information can be sent to multiple literature databases through an application programming interface to obtain a set of candidate documents.

[0033] In a practical implementation, prompt word templates can be constructed first, and then the elements in the structured research framework can be expanded into corresponding subject words and free words through a large language model. The search query can be generated based on the logical combination of the subject words and free words.

[0034] It should be understood that subject terms refer to a standardized dictionary of terms officially maintained by the bibliographic database. Each subject term represents a clear and standardized concept used to index bibliographic content, unaffected by the author's preferred terminology. A subject term can cover all synonyms, near-synonyms, and different word forms expressing the same concept. Using subject terms for retrieval ensures high recall and avoids missing relevant documents due to vocabulary variations.

[0035] Free terms refer to natural language words that appear in the title, abstract, and keywords of a document and are not officially standardized by the document database. They include synonyms, abbreviations, shortened forms, and different spellings. Free terms are non-standardized, flexible, and entirely dependent on the wording of the original author. Free terms can supplement subject term searches, used to find new concepts or specific expressions that have not yet been included as subject terms, thereby improving precision.

[0036] It should be explained that a search query can refer to a query command that can be recognized and executed by a literature database, formed by logically combining subject terms and free terms according to search dimensions (P, I, C, O, S) using Boolean logic operators (such as AND, OR, and NOT) and specific syntax.

[0037] It should be understood that different literature databases have different search syntaxes and API interface specifications, meaning that a directly generated search query cannot be directly sent to multiple literature databases. Therefore, converting the search query into a standard query format enables the standardization and automated delivery of search instructions, thereby achieving efficient and batch cross-database searches.

[0038] Understandably, the candidate document set can refer to the collection of document data that meets the search criteria, which the system initially obtains from multiple document databases by executing a search query.

[0039] Step S30: Extract information from the candidate literature set to obtain quantitative research data, and perform meta-analysis based on the quantitative research data to obtain the meta-analysis results.

[0040] In its implementation, the system first retrieves and parses the full text of documents from the candidate document set. It can then obtain the publisher's address via https: / / doi.org / based on the Document Digital Object Identifier (DOI) from the candidate document set. If the user has download permissions, the system scans the publisher's page to download the full text and names the downloaded file with the DOI. If the user lacks download permissions, the system outputs the document information, allowing the user to manually download the document. Then, it parses all PDF files in the user-specified path, reads the DOIs, consolidates the PDF files in the working directory, and renames them with their respective DOIs. The system can utilize a PDF parsing library to convert PDF files into machine-readable plain text, overcoming the format barriers of PDFs.

[0041] It should be noted that the system can construct prompt word templates for information extraction, and then use a large language model to scan and parse the full text of the literature to extract quantitative research data based on the prompt word templates.

[0042] In practical implementation, a large language model can be used to back-substitute and validate each piece of quantitative research data, checking whether information such as follow-up, group, and sample size matches the information in the full text. A kernel function can be used to calculate the inner product between the extracted quantitative research data and the literature results; if the inner product is 0, relevant information is returned for manual review. This automates and automates the data validation process, significantly improving the accuracy and reliability of data extraction, thus providing a solid data quality foundation for the statistical results of subsequent meta-analysis and improving overall research efficiency.

[0043] It should be understood that quantitative research data can refer to numerical data extracted from each article in the candidate literature collection that can be directly used for statistical merging.

[0044] In a specific implementation, multiple effect sizes can be determined based on the quantitative research data, and heterogeneity tests can be performed on each effect size to obtain the test results. Based on the test results, a corresponding analytical model can be selected, and the multiple effect sizes can be combined using the analytical model to obtain the pooled effect size. Publication bias detection and subgroup analysis can be performed on the multiple effect sizes to obtain publication bias detection results and subgroup analysis results. The test results, the pooled effect size, the publication bias detection results, and the subgroup analysis results are used as the meta-analysis results.

[0045] It should be understood that effect size is a standardized, quantitative indicator used to measure the strength or magnitude of the association between an intervention and its outcome in a study. It unifies the results of different studies onto a common scale for pooling and comparison. For example, for binary data (e.g., effective / ineffective): odds ratio (OR), hazard ratio (RR), and risk difference (RD) are commonly used; for continuous data (e.g., blood pressure values, scale scores): mean difference (MD) or standardized mean difference (SMD) is commonly used; for survival analysis data: risk ratio (HR) is commonly used.

[0046] It should be noted that heterogeneity testing is used to assess the degree of difference between the results of individual studies that are being combined. This difference may stem from differences in study subjects, interventions, study designs, or measurement methods. Heterogeneity can be tested for each effect size using methods such as the Q test or the I² statistic.

[0047] It's important to explain that an analytical model is a statistical framework used in meta-analysis to calculate the pooled effect size and its confidence interval. It defines how weights are assigned to each study. Analytical models include fixed-effects models and random-effects models.

[0048] The fixed effects model assumes that all studies are estimating a common, true effect size, and that differences (heterogeneity) between studies are entirely due to sampling error (chance). Therefore, it gives greater weight to studies with larger sample sizes (smaller variance).

[0049] The random effects model assumes that each study estimates a different true effect size, but one that is distributed around a mean effect size. It acknowledges the existence of genuine heterogeneity among studies. Therefore, when assigning weights, it considers not only sampling error within studies but also variance due to heterogeneity among studies.

[0050] In practice, if the heterogeneity test results indicate that the heterogeneity among studies is small or insignificant, a fixed-effects model should be preferred. If the heterogeneity test results indicate that there is significant heterogeneity among studies, a random-effects model should be chosen.

[0051] Accordingly, the pooled effect size refers to the single overall effect size obtained by weighting all individual effect sizes (i.e., multiple effect sizes) included in the study using the selected analytical model.

[0052] It should be understood that publication bias refers to the tendency for studies with statistically significant positive results to be published more often than those with negative results, which can lead to meta-analyses overestimating the true effect of an intervention. Publication bias detection identifies the presence of this bias. Subgroup analysis is used to investigate potential sources of heterogeneity and answer the question of whether the effect of an intervention differs across studies or populations with different characteristics.

[0053] In the specific implementation, after obtaining the meta-analysis results, the process also includes: generating visualization graphics files and text files that conform to the reporting specifications based on the meta-analysis results; packaging the visualization graphics files and text files into a downloadable document package and providing a version identifier to enable result traceability.

[0054] This embodiment discloses receiving a natural language research question input by a user, parsing the natural language research question, generating a structured research framework based on the parsing results, constructing a search query based on the structured research framework, retrieving literature from multiple literature databases using the search query to obtain a candidate literature set, extracting information from the candidate literature set to obtain quantitative research data, and performing meta-analysis based on the quantitative research data to obtain meta-analysis results. Because this embodiment parses the natural language research question, generates a structured research framework based on the parsing results, then constructs a search query based on the structured research framework to retrieve literature from multiple literature databases, and finally extracts information from the candidate literature set and performs meta-analysis based on the extracted quantitative research data, compared to existing technologies, this embodiment achieves end-to-end automation, improving the efficiency and accuracy of meta-analysis.

[0055] refer to Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the AI-based meta-analysis method of the present invention.

[0056] Based on the first embodiment described above, in this embodiment, step S30 includes steps S301 to S304: Step S301: Extract the title and abstract information of each document in the candidate document set.

[0057] Step S302: Based on the title and abstract information, determine whether the corresponding documents in the candidate document set meet the preset inclusion criteria using a large language model.

[0058] Step S303: Select the candidate literature set that meets the preset inclusion criteria as the target literature set.

[0059] Step S304: Extract information from the target literature set to obtain quantitative research data, and perform meta-analysis based on the quantitative research data to obtain the meta-analysis results.

[0060] It should be understood that pre-defined inclusion criteria can refer to structured rules based on a structured research framework, used to objectively determine whether a document is suitable for inclusion in the final meta-analysis. By screening the candidate document set through pre-defined inclusion criteria, the document screening process is transformed from relying on researchers' personal experience and subjective judgment into a repeatable, traceable, automated, and intelligent decision-making process, thereby greatly improving the efficiency and objectivity of meta-analysis.

[0061] Understandably, the information provided in titles and abstracts is often highly generalized. Therefore, screening based on titles and abstracts (initial screening) is an efficient "coarse screening" process. This is because titles and abstracts typically provide highly generalized information; abstracts often focus on results and conclusions, while the description of the research design may not be detailed enough; some abstracts may exaggerate positive results or omit data that do not meet the pre-specified outcome indicators. Therefore, screening based solely on titles and abstracts carries the risk of including ineligible literature (false positives) or excluding compliant literature (false negatives).

[0062] To improve the accuracy of screening and reduce the risk of incorrect inclusion, after the step of using the documents in the candidate document set that meet the preset inclusion criteria as the target document set, the method further includes: obtaining the text data of each document in the target document set; based on the text data, determining whether the corresponding document in the target document set meets the preset inclusion criteria using the large language model; and using the documents in the target document set that meet the preset inclusion criteria as the screened target document set.

[0063] It should be understood that screening based on text data (secondary screening) is precisely to overcome the limitations of the initial screening. As a "fine screening" process, it further improves the accuracy of screening and reduces the risk of incorrect inclusion.

[0064] Understandably, the purpose of the initial screening is to quickly eliminate a large number of obviously irrelevant documents with minimal computational cost, thereby avoiding the costly process of acquiring and parsing the full text of massive amounts of literature. While the secondary screening based on the text data sacrifices a small amount of efficiency (requiring the acquisition and parsing of the full text), it greatly improves the reliability of the screening results.

[0065] This embodiment discloses the following steps: extracting the title and abstract information of each document in the candidate document set; determining whether the corresponding document in the candidate document set meets the preset inclusion criteria using a large language model based on the title and abstract information; selecting the documents in the candidate document set that meet the preset inclusion criteria as the target document set; obtaining the text data of each document in the target document set; determining whether the corresponding document in the target document set meets the preset inclusion criteria using the large language model based on the text data; selecting the documents in the target document set that meet the preset inclusion criteria as the filtered target document set; extracting information from the target document set to obtain quantitative research data; and performing meta-analysis based on the quantitative research data to obtain meta-analysis results. Because this embodiment performs initial screening of documents in the candidate document set based on title and abstract information using a large language model, and then performs secondary screening of documents in the target document set based on text data using a large language model, compared to existing technologies, this embodiment not only improves screening efficiency but also enhances the reliability of the screening results.

[0066] refer to Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the AI-based meta-analysis method of the present invention.

[0067] Based on the above embodiments, in this embodiment, after step S20, steps S211 to S214 are further included: Step S211: Perform deduplication on the candidate document set to obtain a deduplicated candidate document set, and extract the title and abstract information of each document in the deduplicated candidate document set.

[0068] Step S212: Use the title and abstract information of each document in the deduplicated candidate document set as the information to be evaluated.

[0069] Step S213: Convert the information to be evaluated into a corresponding document vector, and convert the structured research framework into a structured research framework vector.

[0070] Step S214: Determine the similarity between each document vector and the structured research framework vector, and conduct a feasibility assessment of the deduplicated candidate document set based on the similarity to obtain the assessed candidate document set.

[0071] It should be noted that you can first extract information such as journal, author, publication date, title, abstract, keywords, and DOI from the candidate literature set, convert it into structured data, and then perform deduplication. It can also output the number of documents removed due to duplication.

[0072] It should be understood that the similarity between each document vector and the structured research framework vector can be either cosine similarity or Euclidean distance. This embodiment uses cosine similarity as an example. The higher the cosine similarity value (the closer to 1), the stronger the relevance between the document vector and the research question.

[0073] In practice, if the maximum value of each similarity score is less than a preset similarity threshold, it indicates that the documents in the deduplicated candidate document set deviate significantly from the research intent, and the meta-analysis may be infeasible. The system can issue a warning to the user, suggesting that they revise or refine their research question, thus avoiding subsequent ineffective and costly automated processes. Conversely, if the maximum value of each similarity score is greater than the preset similarity threshold, the meta-analysis is deemed feasible.

[0074] This embodiment discloses a method for deduplicating the candidate literature set to obtain a deduplicated candidate literature set, and extracting the title and abstract information of each document in the deduplicated candidate literature set; using the title and abstract information of each document in the deduplicated candidate literature set as evaluation information; converting the evaluation information into corresponding document vectors, and converting the structured research framework into a structured research framework vector; determining the similarity between each document vector and the structured research framework vector, and performing a feasibility assessment on the deduplicated candidate literature set based on the similarity to obtain an evaluated candidate literature set. Because this embodiment deduplicatizes the candidate literature set and performs a feasibility assessment on the deduplicated candidate literature set to obtain an evaluated candidate literature set, compared to existing technologies, this embodiment pre-assesses the feasibility of the candidate literature set before investing significant computational resources in fine-tuning, thus avoiding resource waste.

[0075] Furthermore, this embodiment of the invention also proposes a storage medium storing an AI-based meta-analysis program, which, when executed by a processor, implements the steps of the AI-based meta-analysis method described above.

[0076] Reference Figure 4 , Figure 4 This is a structural block diagram of the first embodiment of the AI-based Meta-analysis system of the present invention.

[0077] like Figure 4 As shown, the artificial intelligence-based meta-analysis system proposed in this embodiment of the invention includes: a problem analysis module 501, a literature screening module 502, and an information extraction module 503.

[0078] The problem parsing module 501 is used to receive natural language research questions input by the user, parse the natural language research questions, and generate a structured research framework based on the parsing results.

[0079] The literature screening module 502 is used to construct a search formula based on the structured research framework, and use the search formula to retrieve literature from multiple literature databases to obtain a candidate literature set.

[0080] The information extraction module 503 is used to extract information from the candidate literature set, obtain quantitative research data, and perform meta-analysis based on the quantitative research data to obtain meta-analysis results.

[0081] The question parsing module 501 is further configured to receive a natural language research question input by a user and obtain context information corresponding to the natural language research question; perform multi-turn conversational parsing on the natural language research question based on the context information to obtain parsing results; determine the key information of the natural language research question based on the parsing results, and generate a structured research framework based on the key information.

[0082] The document screening module 502 is further configured to expand each element in the structured research framework into corresponding subject terms and free terms through a large language model; generate a search query based on the logical combination of the subject terms and free terms, and convert the search query into a query request information in a standard query format; and send the query request information to multiple document databases through an application programming interface to obtain a candidate document set.

[0083] The information extraction module 503 is further configured to determine multiple effect sizes based on the quantitative research data, and perform heterogeneity tests on each effect size to obtain test results; select a corresponding analysis model based on the test results, and use the analysis model to merge the multiple effect sizes to obtain a pooled effect size; perform publication bias detection and subgroup analysis based on the multiple effect sizes to obtain publication bias detection results and subgroup analysis results; and use the test results, the pooled effect size, the publication bias detection results, and the subgroup analysis results as the meta-analysis results.

[0084] This system embodiment discloses receiving a natural language research question input by a user, parsing the natural language research question, generating a structured research framework based on the parsing results, constructing a search query based on the structured research framework, retrieving literature from multiple literature databases using the search query to obtain a candidate literature set, extracting information from the candidate literature set to obtain quantitative research data, and performing meta-analysis based on the quantitative research data to obtain meta-analysis results. Because this system embodiment parses the natural language research question, generates a structured research framework based on the parsing results, then constructs a search query based on the structured research framework to retrieve literature from multiple literature databases, and finally extracts information from the candidate literature set and performs meta-analysis based on the extracted quantitative research data, compared to existing technologies, this system embodiment achieves end-to-end automation, improving the efficiency and accuracy of meta-analysis.

[0085] Based on the first embodiment of the AI-based meta-analysis system of the present invention described above, a second embodiment of the AI-based meta-analysis system of the present invention is proposed.

[0086] In this embodiment, the literature screening module 502 is further configured to extract the title and abstract information of each document in the candidate literature set; based on the title and abstract information, determine whether the corresponding document in the candidate literature set meets the preset inclusion criteria using a large language model; take the documents in the candidate literature set that meet the preset inclusion criteria as the target literature set; extract information from the target literature set to obtain quantitative research data, and perform meta-analysis based on the quantitative research data to obtain meta-analysis results.

[0087] The document screening module 502 is further configured to obtain the text data of each document in the target document set; based on the text data, determine whether the corresponding document in the target document set meets the preset inclusion criteria through the large language model; and use the documents in the target document set that meet the preset inclusion criteria as the screened target document set.

[0088] Other embodiments or specific implementations of the artificial intelligence-based meta-analysis system of the present invention can be found in the above-described method embodiments, and will not be repeated here.

[0089] This application provides an AI-based meta-analysis device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the AI-based meta-analysis method described in Embodiment 1 above.

[0090] The following is for reference. Figure 5 This document illustrates a structural schematic diagram of an AI-based meta-analysis device suitable for implementing embodiments of this application. The AI-based meta-analysis device in these embodiments may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The AI-based meta-analysis device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments in this application.

[0091] like Figure 5 As shown, an AI-based meta-analysis device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes based on programs stored in read-only memory 1002 or programs loaded from storage device 1003 into random access memory 1004. Random access memory 1004 also stores various programs and data required for the operation of the AI-based meta-analysis device. The processing unit 1001, read-only memory 1002, and random access memory 1004 are interconnected via bus 1005. Input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the AI-based meta-analysis device to communicate wirelessly or wiredly with other devices to exchange data. While the figure shows AI-based meta-analysis devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0092] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0093] The AI-based meta-analysis device provided in this application employs the AI-based meta-analysis method described in the above embodiments, which solves the technical problems of low efficiency and insufficient accuracy of traditional meta-analysis methods. Compared with the prior art, the beneficial effects of the AI-based meta-analysis device provided in this application are the same as those of the AI-based meta-analysis method provided in the above embodiments, and other technical features of this AI-based meta-analysis device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0094] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0095] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0096] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0097] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0099] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the protection scope of the present invention.

Claims

1. A meta-analysis method based on artificial intelligence, characterized in that, The method includes: The system receives natural language research questions input by users, parses the natural language research questions, and generates a structured research framework based on the parsing results. Based on the structured research framework, a search formula is constructed, and the search formula is used to retrieve literature from multiple literature databases to obtain a candidate literature set; Information is extracted from the candidate literature set to obtain quantitative research data, and meta-analysis is performed based on the quantitative research data to obtain the meta-analysis results.

2. The meta-analysis method based on artificial intelligence as described in claim 1, characterized in that, The steps of extracting information from the candidate literature set to obtain quantitative research data, and performing meta-analysis based on the quantitative research data to obtain meta-analysis results include: Extract the title and abstract information of each document in the candidate document set; Based on the title and abstract information, a large language model is used to determine whether the corresponding documents in the candidate document set meet the preset inclusion criteria. The literature that meets the preset inclusion criteria in the candidate literature set is taken as the target literature set; Information is extracted from the target literature set to obtain quantitative research data, and meta-analysis is performed based on the quantitative research data to obtain the meta-analysis results.

3. The meta-analysis method based on artificial intelligence as described in claim 2, characterized in that, After the step of selecting the candidate literature set that meets the preset inclusion criteria as the target literature set, the method further includes: Obtain the text data of each document in the target document set; Based on the text data, the large language model is used to determine whether the corresponding documents in the target document set meet the preset inclusion criteria. The literature that meets the preset inclusion criteria in the target literature set is used as the filtered target literature set.

4. The meta-analysis method based on artificial intelligence as described in claim 1, characterized in that, After the step of retrieving documents from multiple literature databases using the search formula to obtain a candidate document set, the method further includes: The candidate document set is deduplicated to obtain a deduplicated candidate document set, and the title and abstract information of each document in the deduplicated candidate document set are extracted. The title and abstract information of each document in the deduplicated candidate document set are used as the information to be evaluated; The information to be evaluated is converted into a corresponding document vector, and the structured research framework is converted into a structured research framework vector. The similarity between each document vector and the structured research framework vector is determined, and the feasibility of the deduplicated candidate document set is evaluated based on the similarity to obtain the evaluated candidate document set.

5. The meta-analysis method based on artificial intelligence as described in claim 1, characterized in that, The steps of constructing a search query based on the structured research framework, and using the search query to retrieve literature from multiple literature databases to obtain a candidate literature set include: The elements in the structured research framework are expanded into corresponding topic words and free words using a large language model; A search query is generated based on the logical combination of the subject terms and free terms, and the search query is converted into a query request information in a standard query format. The query request information is sent to multiple literature databases through the application programming interface to obtain a set of candidate documents.

6. The meta-analysis method based on artificial intelligence as described in claim 1, characterized in that, The steps for performing meta-analysis based on the quantitative research data to obtain meta-analysis results include: Based on the quantitative research data, multiple effect sizes are determined, and heterogeneity tests are performed on each effect size to obtain the test results; Based on the test results, a corresponding analytical model is selected, and the analytical model is used to combine multiple effect sizes to obtain the combined effect size. Publication bias detection and subgroup analysis were performed based on the various effect sizes to obtain publication bias detection results and subgroup analysis results; The test results, the pooled effect size, the publication bias detection results, and the subgroup analysis results were used as the results of the meta-analysis.

7. The meta-analysis method based on artificial intelligence as described in any one of claims 1-6, characterized in that, The steps of receiving a natural language research question input by a user, parsing the natural language research question, and generating a structured research framework based on the parsing result include: Receive a natural language research question input by the user and obtain the context information corresponding to the natural language research question; Based on the context information, a multi-turn conversational parsing is performed on the natural language research question to obtain the parsing results; Based on the parsing results, key information about the natural language research problem is determined, and a structured research framework is generated based on the key information.

8. A meta-analysis system based on artificial intelligence, characterized in that, The system includes: The problem parsing module is used to receive natural language research questions input by users, parse the natural language research questions, and generate a structured research framework based on the parsing results. The literature screening module is used to construct a search query based on the structured research framework, and use the search query to retrieve literature from multiple literature databases to obtain a candidate literature set; The information extraction module is used to extract information from the candidate literature set, obtain quantitative research data, and perform meta-analysis based on the quantitative research data to obtain meta-analysis results.

9. A meta-analysis device based on artificial intelligence, characterized in that, The device includes: a memory, a processor, and an AI-based meta-analysis program stored in the memory and executable on the processor, the AI-based meta-analysis program being configured to implement the steps of the AI-based meta-analysis method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores an AI-based meta-analysis program, which, when executed by a processor, implements the steps of the AI-based meta-analysis method as described in any one of claims 1 to 7.

Images