GPT-based scientific research hotspot prediction method and device

Abstract

The invention discloses a GPT-based scientific research hotspot prediction method and device, and belongs to the technical field of natural language processing. The method comprises the following steps: collecting scientific research literatures in a public database, and cleaning, classifying and labeling the scientific research literatures to generate a data set; the keywords appearing in the scientific research literature in the data set are sorted and stored according to the appearing times; performing initial setting on the GPT model, and providing a data set for the GPT model to train the GPT model; predicting the scientific research hotspot keyword by adopting the trained GPT model to obtain a prediction result; and generating a scientific research hotspot prediction report based on the prediction result. According to the provided historical data, the keywords of the field in the future can be predicted, the preliminary hotspot prediction result is generated, and the prediction result is cycled for multiple times and comprehensively analyzed, so that the finally obtained scientific research hotspots are more accurate. A powerful tool is provided for scientific researchers, and the efficiency and quality of scientific research activities are greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of natural language processing, and specifically relates to a method and device for predicting scientific research hotspots based on GPT. Background Art

[0002] In modern scientific research activities, understanding and predicting future research hotspots is an important part of scientific research and technological innovation. Traditional research hotspot prediction methods mainly rely on expert review and bibliometric analysis. However, these methods have some obvious defects and shortcomings:

[0003] (1) Subjectivity of expert review: Expert review relies on the personal experience and judgment of experts, which is highly subjective and limited. Different experts may have different evaluation criteria, resulting in inconsistent prediction results. In addition, the number of experts is limited, and the fields covered are relatively narrow, making it difficult to fully reflect the research dynamics of the entire discipline.

[0004] (2) The lag of bibliometric analysis: Bibliometric analysis predicts research hotspots by counting and analyzing the number of published documents and their citations. This method usually requires a large amount of historical data, and the data collection and processing process is cumbersome and time-consuming. Since there is a certain time lag in the publication and citation of documents, the timeliness and accuracy of the prediction results are limited.

[0005] (3) Complexity of data processing: In practical applications, the amount of scientific research literature data is huge and the structure is complex, involving multiple data types and formats, such as text, charts, references, etc. Traditional data processing and analysis methods are difficult to efficiently process and analyze such a large amount of data, and lack the ability to deeply understand and mine the complex relationships between data.

[0006] (4) Limitations of trend analysis: Existing research hotspot prediction methods are mainly based on the statistics and trend analysis of historical data, which cannot fully consider the dynamic changes and innovation trends in the field of scientific research. With the continuous emergence of new technologies and new methods, the pace of change in research hotspots has accelerated, and traditional methods are difficult to capture and predict these changes in a timely and accurate manner. Summary of the invention

[0007] The present invention aims to address the deficiencies of the prior art and proposes a scientific research hotspot prediction method and device based on GPT. By utilizing the powerful natural language processing and generation capabilities of the generative pre-training model (GPT), a large amount of historical scientific research literature data is deeply analyzed and mined to accurately predict future scientific research hotspots.

[0008] To achieve the above object, the present invention provides the following scheme: a method for predicting scientific research hotspots based on GPT, comprising the following steps:

[0009] S1. Collect scientific research documents from public databases, clean, classify and annotate them to generate a data set; and sort and save the keywords appearing in the scientific research documents in the data set according to the number of occurrences;

[0010] S2. Initially set the GPT model and provide the data set to the GPT model to train the GPT model;

[0011] S3. Use the trained GPT model to predict the scientific research hotspot keywords to obtain prediction results; and generate a scientific research hotspot prediction report based on the prediction results.

[0012] Further preferably, S2 comprises the following steps:

[0013] S21. Design a prompting project, which is used to ensure that the GPT model understands the analysis context and purpose;

[0014] S22, providing the data set to the GPT model, and displaying the sorted keywords according to the number of occurrences; the GPT model analyzes the keywords to obtain training results;

[0015] S23. Evaluate the predictive ability of the trained GPT model based on the accuracy and repetition rate of the training results.

[0016] Further preferably, S3 comprises the following steps:

[0017] S31. Use the trained GPT model to predict the scientific research hotspot keywords and obtain the original prediction results;

[0018] S32, inputting the original prediction result as a new input into the trained GPT model to obtain a secondary prediction result;

[0019] S33, setting a cycle threshold, repeating the S32 process until the cycle ends; keeping the result of each cycle to obtain the prediction result;

[0020] S34. Analyze the prediction results to obtain scientific research hotspots and generate a scientific research hotspot prediction report.

[0021] Further preferably, the accuracy calculation method includes:

[0022] Accuracy = (number of correctly predicted words) / (number of original words provided) × 100%;

[0023] The calculation method of the repetition rate includes:

[0024] Repetition rate = (number of predicted words that match the provided word) / (number of original words provided) x 100%.

[0025] The present invention also provides a scientific research hotspot prediction device based on GPT, comprising: a data set acquisition system, a model training system and a prediction system;

[0026] The data set collection system is used to collect scientific research documents from public databases, and clean, classify and annotate the scientific research documents to generate data sets; and sort and save the keywords appearing in the scientific research documents in the data sets according to the number of occurrences;

[0027] The model training system is used to perform initial settings on the GPT model and provide the data set to the GPT model to train the GPT model;

[0028] The prediction system is used to predict scientific research hotspot keywords using the trained GPT model to obtain prediction results; and generate a scientific research hotspot prediction report based on the prediction results.

[0029] Further preferably, the model training system comprises: a design module, a training module and an evaluation module;

[0030] The design module is used to design a prompting project, and the prompting project is used to ensure that the GPT model understands the analysis context and purpose;

[0031] The training module is used to provide the data set to the GPT model and display the sorted keywords according to the number of occurrences; the GPT model analyzes the keywords to obtain training results;

[0032] The evaluation module is used to evaluate the predictive ability of the trained GPT model based on the accuracy and repetition rate of the training results.

[0033] Further preferably, the prediction system comprises: an original prediction module, a circulation module, a storage module and an analysis module;

[0034] The original prediction module is used to predict the scientific research hotspot keywords using the trained GPT model to obtain the original prediction results;

[0035] The loop module is used to input the original prediction result as a new input into the trained GPT model to obtain a secondary prediction result;

[0036] The storage module sets a loop threshold, takes the last prediction result of the loop module as a new input of the loop module, and performs a loop process until the loop ends to obtain the prediction result;

[0037] The analysis module is used to analyze the prediction results, obtain scientific research hotspots, and generate a scientific research hotspot prediction report.

[0038] Further preferably, the accuracy calculation method includes:

[0039] Accuracy = (number of correctly predicted words) / (number of original words provided) × 100%;

[0040] The calculation method of the repetition rate includes:

[0041] Repetition rate = (number of predicted words that match the provided word) / (number of original words provided) x 100%.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention uses the GPT model to automatically process and analyze large-scale scientific research literature data, which can quickly and accurately extract and identify key information, and improve the efficiency of data processing and analysis. Through multiple cyclic predictions and comprehensive analysis of the trends and patterns of historical data, it can timely and accurately capture and predict the dynamic changes of scientific research hotspots, providing powerful decision-making support for scientific researchers and institutions. This method is applicable to the prediction of scientific research hotspots in multiple disciplines and has broad application prospects and value. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0045] Figure 1 A schematic diagram of a process flow of a scientific research hotspot prediction method based on GPT according to an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the GPT model training process according to an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of a situation in which the GPT model of an embodiment of the present invention obtains predicted keywords based on provided keywords. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Embodiment 1:

[0051] like Figure 1 As shown, a scientific research hotspot prediction method based on GPT includes the following steps:

[0052] S1. Collect scientific research literature from public databases, clean, classify and annotate the scientific research literature to generate a data set; and sort and save the keywords appearing in the scientific research literature in the data set according to the number of occurrences.

[0053] This example collects all papers published in the journal Environmental Science & Technology from 2000 to 2019, and exports the retrieved records as plain text files, including the full record and cited references. The keywords that appear most frequently in all documents of the journal Environmental Science & Technology are counted each year.

[0054] S2. Initialize the GPT model and provide the GPT model with a data set to train the GPT model.

[0055] Furthermore, if Figure 2 As shown, S2 includes the following steps:

[0056] S21. Design a prompt project. The prompt project is used to inform GPT of the source and training purpose of the training text, ensuring that the GPT model understands the analysis background and purpose.

[0057] S22. Provide standardized data to the GPT model, display the keywords that appear most frequently in a specific year, and analyze this data through the model. Require the GPT model to predict keywords that may appear in the future environmental field based on the provided historical data, trends, and patterns. Repeat the data provision and hotspot prediction process, conduct multiple cycles of prediction, and comprehensively analyze the results of each prediction. Conduct a comprehensive analysis of the saved prediction results to obtain the most likely predicted keywords. For example, provide keywords and occurrence times from 2000 to 2003 to predict keywords for 2004; provide keywords and occurrence times from 2005 to 2008 to predict keywords for 2009; provide keywords and occurrence times from 2010 to 2013 to predict keywords for 2014; provide keywords and occurrence times from 2015 to 2018 to predict keywords for 2019; Figure 3 As shown, the provided keywords, predicted keywords, and actual words are displayed.

[0058] S23. Evaluate the predictive ability of the trained GPT model based on the accuracy and repetition rate of the training results.

[0059] The accuracy calculation method includes:

[0060] Accuracy = (number of correctly predicted words) / (number of original words provided) × 100%;

[0061] The calculation method of repetition rate includes:

[0062] Repetition rate = (number of predicted words that match the provided word) / (number of original words provided) x 100%.

[0063] Table 1 shows the prediction accuracy and repetition rate in different time periods.

[0064] Table 1

[0065]

[0066] S3. Use the trained GPT model to predict the scientific research hotspot keywords and obtain the prediction results. Generate a scientific research hotspot prediction report based on the prediction results.

[0067] Further, S3 includes the following steps:

[0068] S31. Use the trained GPT model to predict the scientific research hot keywords and obtain the original prediction results.

[0069] S32: The original prediction result is used as a new input into the trained GPT model to obtain a secondary prediction result.

[0070] S33, set the loop threshold, repeat S32 process until the loop ends; keep the result of each loop to obtain the prediction result.

[0071] S34. Analyze the prediction results, obtain scientific research hotspots, and generate a scientific research hotspot prediction report.

[0072] Embodiment 2:

[0073] This embodiment provides a scientific research hotspot prediction device based on GPT, including: a data set acquisition system, a model training system and a prediction system.

[0074] The data set collection system is used to collect scientific research literature from public databases, and to clean, classify and annotate the scientific research literature to generate data sets; and to sort and save the keywords that appear in the scientific research literature in the data set according to the number of occurrences.

[0075] This example collects all papers published in the journal Environmental Science & Technology from 2000 to 2019, and exports the retrieved records as plain text files, including the full record and cited references. The keywords that appear most frequently in all documents of the journal Environmental Science & Technology are counted each year.

[0076] The model training system is used to perform initial settings for the GPT model and provide the GPT model with a data set for training.

[0077] Furthermore, the model training system includes: a design module, a training module and an evaluation module.

[0078] The design module is used to design the prompt project, which is used to inform GPT of the source and training purpose of the training text to ensure that the GPT model understands the analysis background and purpose.

[0079] The training module is used to provide standardized data to the GPT model, display the most frequently appearing keywords in a specific year, and analyze this data through the model. The GPT model is required to predict the keywords that may appear in the future environment field based on the provided historical data, trends and patterns. Repeat the data provision and hotspot prediction process, perform multiple cycles of prediction, and comprehensively analyze the results of each prediction. Comprehensively analyze the saved prediction results to obtain the most likely predicted keywords.

[0080] The evaluation module is used to evaluate the predictive ability of the trained GPT model based on the accuracy and repetition rate of the training results.

[0081] The accuracy calculation method includes:

[0082] Accuracy = (number of correctly predicted words) / (number of original words provided) × 100%;

[0083] The calculation method of repetition rate includes:

[0084] Repetition rate = (number of predicted words that match the provided word) / (number of original words provided) x 100%.

[0085] The prediction system is used to use the trained GPT model to predict scientific research hotspot keywords and obtain prediction results; and generate a scientific research hotspot prediction report based on the prediction results.

[0086] Furthermore, the prediction system includes: an original prediction module, a loop module, a storage module and an analysis module. The original prediction module is used to use the trained GPT model to predict the scientific research hotspot keywords and obtain the original prediction results. The loop module is used to input the original prediction results as new inputs into the trained GPT model to obtain secondary prediction results. The storage module sets the loop threshold, uses the previous prediction results of the loop module as the new input of the loop module, and performs a loop process until the end of the loop; the results of each loop are maintained to obtain the prediction results. The analysis module is used to analyze the prediction results, obtain scientific research hotspots, and generate a scientific research hotspot prediction report.

[0087] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A scientific research hotspot prediction method based on GPT, characterized in that: The following steps are involved: S1. Collect scientific research literature from public databases, clean, classify and annotate them, and generate data sets; And sort and save the keywords appearing in the scientific research documents in the data set according to the number of occurrences; S2. Initially set the GPT model and provide the data set to the GPT model to train the GPT model; S3. Use the trained GPT model to predict the scientific research hotspot keywords to obtain prediction results; and generate a scientific research hotspot prediction report based on the prediction results.

2. According to claim 1, a GPT-based scientific research hotspot prediction method is characterized in that: S2 includes the following steps: S21. Design a prompting project, which is used to ensure that the GPT model understands the analysis context and purpose; S22, providing the data set to the GPT model, and displaying the sorted keywords according to the number of occurrences; The GPT model analyzes keywords and obtains training results; S23. Evaluate the predictive ability of the trained GPT model based on the accuracy and repetition rate of the training results.

3. According to the GPT-based scientific research hotspot prediction method of claim 1, it is characterized in that: S3 includes the following steps: S31. Use the trained GPT model to predict the scientific research hotspot keywords and obtain the original prediction results; S32, inputting the original prediction result as a new input into the trained GPT model to obtain a secondary prediction result; S33, setting a cycle threshold, repeating the S32 process until the cycle ends; keeping the result of each cycle to obtain the prediction result; S34. Analyze the prediction results to obtain scientific research hotspots and generate a scientific research hotspot prediction report.

4. According to claim 2, a GPT-based scientific research hotspot prediction method is characterized in that: The accuracy calculation method includes: Accuracy = (number of correctly predicted words) / (number of original words provided) × 100%; The calculation method of the repetition rate includes: Repetition rate = (number of predicted words that match the provided word) / (number of original words provided) x 100%.

5. A scientific research hotspot prediction device based on GPT, the device is used to implement the method described in any one of claims 1 to 4, characterized in that: include: Data set acquisition system, model training system, and prediction system; The data set collection system is used to collect scientific research documents from public databases, and clean, classify and annotate the scientific research documents to generate data sets; And sort and save the keywords appearing in the scientific research documents in the data set according to the number of occurrences; The model training system is used to perform initial settings on the GPT model and provide the data set to the GPT model to train the GPT model; The prediction system is used to predict scientific research hotspot keywords using the trained GPT model to obtain prediction results; and generate a scientific research hotspot prediction report based on the prediction results.

6. A scientific research hotspot prediction device based on GPT according to claim 5, characterized in that: The model training system includes: design module, training module and evaluation module; The design module is used to design a prompting project, and the prompting project is used to ensure that the GPT model understands the analysis context and purpose; The training module is used to provide the data set to the GPT model and display the sorted keywords according to the number of occurrences; the GPT model analyzes the keywords to obtain training results; The evaluation module is used to evaluate the predictive ability of the trained GPT model based on the accuracy and repetition rate of the training results.

7. A GPT-based scientific research hotspot prediction device according to claim 5, characterized in that: The prediction system includes: an original prediction module, a circulation module, a storage module and an analysis module; The original prediction module is used to predict the scientific research hotspot keywords using the trained GPT model to obtain the original prediction results; The loop module is used to input the original prediction result as a new input into the trained GPT model to obtain a secondary prediction result; The storage module sets a loop threshold, takes the last prediction result of the loop module as a new input of the loop module, and performs a loop process until the loop ends to obtain the prediction result; The analysis module is used to analyze the prediction results, obtain scientific research hotspots, and generate a scientific research hotspot prediction report.

8. A scientific research hotspot prediction device based on GPT according to claim 6, characterized in that: The accuracy calculation method includes: Accuracy = (number of correctly predicted words) / (number of original words provided) × 100%; The calculation method of the repetition rate includes: Repetition rate = (number of predicted words that match the provided word) / (number of original words provided) x 100%.

Images