Retrieval model recommendation method and device based on knowledge graph

Abstract

The invention discloses a retrieval model recommendation method and device based on a knowledge graph, relates to the technical field of artificial intelligence, and solves the technical problem that the quality of retrieval results is uneven due to the fact that retrieval models are relatively single and mostly depend on simple keyword matching. An entity identification and relation extraction technology is used to construct a knowledge graph with a rigorous structure and rich information. Therefore, the relationship between the data can be clearly presented, more comprehensive and accurate data support is provided for the subsequent retrieval model, and the model with the optimal expression can be screened out according to different retrieval problems by performing a series of processes such as labeling, retrieval operation, result analysis and sorting on the retrieval model. According to the comprehensive application and scientific screening mode of multiple models, the limitation of traditional single model retrieval is changed, and the adaptability of the retrieval model and the retrieval problem is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and specifically to a retrieval model recommendation method and device based on a knowledge graph. Background Art

[0002] With the development of artificial intelligence technology, retrieval models are widely used in various fields, such as search engines, natural language processing, etc. Using retrieval models can quickly and efficiently search and analyze a large amount of data.

[0003] According to the application with the publication number CN117540062A, a retrieval model recommendation method and device based on a knowledge graph are disclosed. The method includes: obtaining a plurality of retrieval models and the knowledge graph corresponding to the target platform; analyzing the pre-obtained problem information based on the knowledge graph to obtain a problem analysis result corresponding to the problem information; for each retrieval model, retrieving based on the retrieval model for the problem information to obtain a retrieval result corresponding to the retrieval model; determining the target retrieval model corresponding to the target platform according to the problem analysis result and the retrieval results corresponding to all retrieval models.

[0004] In the traditional retrieval field, the retrieval models are relatively single, mostly relying on simple keyword matching, and it is difficult to accurately understand the complex semantic information and potential technical associations in the text. At the same time, in the face of multi-source heterogeneous data, such as text, drawing information, and examination opinions, there are lack of effective integration and in-depth mining means, and the value of these data cannot be fully utilized. In addition, for different types of retrieval problems, there is no systematic classification and adaptation mechanism, resulting in uneven quality of retrieval results and difficulty in meeting the growing retrieval needs. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention provides a retrieval model recommendation method and device based on a knowledge graph, which solves the problem that the retrieval models are relatively single and mostly rely on simple keyword matching, resulting in uneven quality of retrieval results.

[0006] To achieve the above object, the present invention is realized through the following technical solutions: A retrieval model recommendation method based on a knowledge graph, which specifically includes the following steps: Establish a knowledge graph, obtain all retrieval models and the corresponding retrieval problems, and at the same time determine the problem types corresponding to the retrieval problems; Split the retrieval problem to obtain split information, and at the same time retrieve the split information with different retrieval models respectively to obtain the corresponding retrieval results. Then analyze the content matching degree between the retrieval problem and the retrieval results, and screen the retrieval results based on the content matching degree to obtain preliminary selection results. At the same time, sort the retrieval models according to the number of preliminary selection results to generate model sorting information; Obtain the model sorting information, screen the retrieval question to obtain keywords, and calculate the occurrence frequency of the keywords. Then calculate the semantic similarity between the retrieval question and the retrieval results, and calculate the sum of the values of the occurrence frequency and the semantic similarity to obtain the preselected result association value. Based on the discrete value analysis, obtain the result to be analyzed, and calculate the mean value of the sum of the values of the result to be analyzed association value to obtain the association index; Compare the association index of the retrieval model with the threshold value to screen out the preselected models, and determine the grade assignment of the preselected models based on the retrieval time. At the same time, calculate the sum of its values and the association index to obtain the model benefit value, and select the preselected model with the largest model benefit value to generate a recommended model, and generate recommended information in combination with the retrieval type of the retrieval question.

[0007] As a further solution of the present invention, the specific method for splitting the retrieval question to obtain the split information is as follows: Obtain the retrieval question, and use the bag-of-words model to split the retrieval question. When splitting, the bag-of-words model will decompose the retrieval question into independent words or phrases with domain significance to obtain the corresponding split information.

[0008] As a further solution of the present invention, the specific method for generating the model sorting information is as follows: Obtain all retrieval models and label them as i, where i = 1, 2, …, j, and j represents the number of retrieval models. At the same time, obtain the retrieval result Ci corresponding to the retrieval model i, obtain the split information, and calculate the proportion of the quantity of the split information corresponding to the retrieval result Ci; Compare it with the preset value, and screen out the retrieval results with a proportion greater than the preset value and record them as the preselected results; By analogy, screen the preselected results corresponding to all retrieval models i, and sort them in descending order according to the quantity of the screening results to generate the model sorting information.

[0009] As a further solution of the present invention, the specific method for screening the retrieval question to obtain keywords is as follows: Obtain the retrieval question, and at the same time, with the help of natural language processing tools, analyze the retrieval question according to the lexical rules, and comprehensively process the retrieval question in combination with the stop word filtering method to screen out the keywords corresponding to the retrieval question, and label them as n, where n = 1, 2, …, m, and m represents the number of keywords.

[0010] As a further solution of the present invention, the specific method for calculating the occurrence frequency of the keywords is as follows: Perform sentence splitting on the preselected result a to obtain sentence splitting information, calculate the number of occurrences of the keyword n in the sentence splitting information, and calculate the proportion of the number of occurrences to the total number of the sentence splitting information; And so on, calculate the proportion value corresponding to the keyword n in all clause information, then sum up the calculated proportion values to obtain the occurrence frequency of the keyword n.

[0011] As a further solution of the present invention, the specific method for calculating the mean value of the sum of the associated values of the results to be analyzed to obtain the association index is as follows: Using the word vector model, calculate the semantic similarity between the retrieval problem and the retrieval result, and calculate the sum of the numerical values of the semantic similarity and the occurrence frequency, which is denoted as the preselected result association value. Then, screen the preselected results according to the discrete values to obtain the results to be analyzed. At the same time, calculate the mean value of the sum of the associated values of the results to be analyzed, which is denoted as the association index between the analysis model to be analyzed and the retrieval problem; And so on, calculate the association indexes of all retrieval models, and sort them from large to small according to the association indexes.

[0012] As a further solution of the present invention, the specific method for obtaining the model benefit value is as follows: Obtain the association index corresponding to the retrieval model, compare it with the threshold value, screen the retrieval models with the association index greater than the threshold value, which are denoted as the preselected models. Then, obtain the retrieval time corresponding to the preselected models, and based on big data analysis, divide the retrieval time into grade intervals and perform corresponding assignment processing. At the same time, match the retrieval time of the preselected models with the grade intervals to determine the retrieval time grade corresponding to the preselected models, and obtain the corresponding grade assignment. Then, calculate the sum value of the grade assignment and the association index, which is denoted as the model benefit value.

[0013] As a further solution of the present invention, the specific method for generating the recommended information by combining the retrieval type of the retrieval problem is as follows: Select the preselected model with the largest model benefit value as the standard to generate a recommended model, obtain the corresponding retrieval problem type, and combine it with the recommended model to generate the recommended information; And so on, obtain historical data, and obtain the retrieval types corresponding to all deceleration models to generate the corresponding recommended information.

[0014] A retrieval model recommendation device based on a knowledge graph, comprising: An information acquisition module, configured to acquire a plurality of retrieval models and the corresponding knowledge graphs; An information analysis module, configured to analyze the pre-acquired retrieval problem based on the knowledge graph to obtain the retrieval result corresponding to the retrieval problem; A retrieval analysis module, configured to perform retrieval on the retrieval problem for each of the retrieval models to obtain the retrieval result corresponding to the retrieval model; A retrieval determination module, configured to determine the corresponding target retrieval model according to the retrieval results corresponding to all the retrieval models, and generate the recommended information.

[0015] Beneficial effects The present invention provides a retrieval model recommendation method and device based on a knowledge graph. Compared with the prior art, it has the following beneficial effects: By collecting multi-source heterogeneous data and applying entity recognition and relationship extraction technologies, the present invention constructs a knowledge graph with a rigorous structure and rich information. This clearly presents the relationships between data, providing more comprehensive and accurate data support for subsequent retrieval models. Compared with traditional methods that only rely on a single data source or simple data processing methods, it improves the utilization value and depth of data.

[0016] It covers various retrieval models based on keyword matching, semantic understanding deep learning, and the structure of the knowledge graph. Through a series of processes such as labeling, retrieving, result analyzing, and ranking the retrieval models, the optimal model can be selected according to different retrieval problems. This comprehensive application and scientific screening method of multiple models change the limitations of traditional single-model retrieval and significantly improve the adaptability of the retrieval model to retrieval problems.

[0017] During the retrieval process, through operations such as splitting the retrieval problem, keyword screening, and semantic similarity calculation, combined with multi-dimensional analysis of the retrieval results, such as calculating the quantity ratio, correlation index, etc., and grading and assigning values according to the retrieval time, the recommended model is finally determined. This series of innovative methods greatly improves the matching degree between the retrieval results and the retrieval problem, significantly improves the retrieval efficiency, effectively controls the retrieval time, and provides a faster and more accurate retrieval service for retrieval personnel compared with traditional retrieval methods, helping them obtain the required information more efficiently. Description of the drawings

[0018] Figure 1 It is a flowchart of the method steps of the present invention; Figure 2 It is a block diagram of the device of the present invention. Specific embodiments

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Embodiment 1, please refer to Figure 1 , the present application provides a retrieval model recommendation method based on a knowledge graph, and this method specifically includes the following steps: Step1、 First, start building a knowledge graph. By collecting multi-source heterogeneous data, including text, drawing information, examination opinions, etc., and using entity recognition technology, various entities are accurately extracted from this data, such as the invention name, technical field, classification number, and other entities in it. Then, using relationship extraction technology, determine the association relationships between entities, such as the "invention name - belongs to - technical field" relationship. With the help of knowledge fusion means, integrate the same entity information obtained from different data sources, eliminate data redundancy and conflicts, and finally build a knowledge graph with a rigorous structure and rich information. For example, when building a knowledge graph in the field of electronic communication, entities such as "5G communication technology" and "chip manufacturing process" are extracted from a large number of electronic communication literatures, as well as the relationship "5G communication technology - application - chip manufacturing process" between them. After completing the construction of the knowledge graph, comprehensively obtain all retrieval models applicable to retrieval. These retrieval models cover various types, including traditional retrieval models based on keyword matching, which achieve retrieval by finding content that exactly matches or is fuzzy-matched with the input keyword in the entity and relationship descriptions of the knowledge graph; deep learning retrieval models based on semantic understanding, which use deep learning algorithms to deeply mine and understand the semantic information in the knowledge graph and can handle retrieval problems with complex semantics; and graph retrieval models based on the structure of the knowledge graph, which perform retrieval based on the structural features of nodes and edges in the knowledge graph, such as obtaining relevant information by finding adjacent nodes of a specific node or a specific path.

[0021] Step2、 First, comprehensively sort out and collect various retrieval models applicable to retrieval. These models may cover traditional retrieval models based on keyword matching, deep learning retrieval models based on semantic understanding, and professional retrieval models designed according to the text structure characteristics, etc. Number all the collected retrieval models in sequence, denoted as i, and i = 1, 2, …, j, where j represents the number of retrieval models. For example, if we have collected three models: a model based on title keyword retrieval, a model based on full-text semantic retrieval, and a model based on classification number retrieval, then j = 3, and the corresponding i are 1, 2, and 3 respectively; For each retrieval model i, use it to perform a retrieval operation on the input retrieval question to obtain the corresponding retrieval result Ci. For example, retrieval model 1 is a model based on title keywords. When the retrieval question is "a new type of energy-saving air conditioner", its retrieval result C1 may be a list of documents whose titles contain keywords related to "energy-saving air conditioner". Next, the retrieval question is split. Here, the bag-of-words model is adopted. Due to the professionalism and standardization of the text, the bag-of-words model will decompose the retrieval question into individual independent words or phrases with domain significance, ignoring the order relationship between them. For example, for the retrieval question "find a smartphone with an efficient heat dissipation structure", after being split by the bag-of-words model, the possible split information may be four groups: "find", "efficient heat dissipation structure", "smartphone", "".

[0022] Then, for each retrieval result Ci, calculate the proportion of the corresponding quantity of the split information. The specific calculation method is to carefully count the quantity of the split information contained in the retrieval result Ci and then divide it by the total quantity of the split information. Suppose the split information obtained after splitting is five groups, and the corresponding split information in a certain retrieval result Ci is three groups. Then the proportion of this retrieval result is 3 / 5. After that, compare the obtained proportion with the preset value.

[0023] The specific value of the preset value is set by the retrieval operator according to the actual requirements such as the accuracy requirement of the retrieval target, the retrieval time limit, and past experience. For example, the operator sets the preset value to 0.6 according to the relatively high accuracy expected in this retrieval. Select the retrieval results with a proportion greater than the preset value and record them as preselected results. In this way, process and screen the retrieval results corresponding to all retrieval models i in turn. For example, for retrieval model 1 in the above example, if the calculated proportion is =0.6, exactly equal to the preset value, if the preset value is set to a strictly greater relationship, then this retrieval result C1 is not screened as a preselected result; if the preset value is set to a greater than or equal relationship, then this retrieval result C1 is screened as a preselected result.

[0024] Repeat this process for all retrieval models i (from 1 to j) to complete the screening of the preselected results corresponding to all retrieval models. Finally, sort the selected preselected results in descending order of quantity to generate model sorting information. For example, the number of preselected results screened by retrieval model 2 is 100, the number of preselected results screened by retrieval model 3 is 80, and the number of preselected results screened by retrieval model 1 is 50. Then the generated model sorting information is that retrieval model 2 ranks first, retrieval model 3 ranks second, and retrieval model 1 ranks last. Through such sorting, we can intuitively understand the performance of different retrieval models on this retrieval question, providing a strong basis for subsequent optimization of the retrieval strategy or selection of the retrieval model.

[0025] Step 3、 Obtain the corresponding sorting information, and take one set of retrieval models as the analysis object for analysis. Obtain the corresponding preliminary selection results of the analysis object and label them as a, where a = 1, 2, …, b, and b represents the number of preliminary selection results of the analysis object. At the same time, obtain the retrieval time corresponding to the analysis object, and the retrieval time here refers to the duration of obtaining the retrieval results according to the retrieval problem. Then, perform keyword screening processing on the retrieval problem, and the specific screening processing method is as follows: Obtain the retrieval problem. At the same time, with the help of natural language processing tools, analyze the retrieval problem according to the lexical rules. For example, use a part-of-speech tagging tool to identify words of different parts of speech such as nouns, verbs, and adjectives. Usually, nouns are entity concepts, and verbs represent actions or behaviors, which are relatively key information and can be preferentially considered as keywords. Combine the stop-word filtering method to comprehensively process the retrieval problem. Stop words are words that frequently appear in the text but have little effect on expressing the core meaning, such as "of", "is", "in", "how", etc. By establishing a stop-word list, filter out the stop words in the retrieval problem. The remaining words often better reflect the key content of the problem. Screen out the keywords corresponding to the retrieval problem and label them as n, where n = 1, 2, …, m, and m represents the number of keywords. Then, analyze the number of occurrences of keyword n in the preliminary selection results a, and calculate the occurrence frequency of keyword n in the preliminary selection results a, denoted as Pa. The calculation method of the occurrence frequency is to perform sentence segmentation processing on the preliminary selection results a to obtain sentence segmentation information, calculate the number of occurrences of keyword n in the sentence segmentation information, and at the same time calculate the ratio of the number of occurrences to the total number of sentence segmentation information. By analogy, calculate the ratio corresponding to keyword n in all sentence segmentation information, and then sum up the calculated ratios to obtain the occurrence frequency of keyword n; For example, first obtain the retrieval problem, such as "search for the manufacturing process of lithium-ion batteries with high safety and long life". With the help of professional natural language processing tools, analyze it according to the lexical rules. Take the part-of-speech tagging tool as an example. By analysis, words of different parts of speech can be identified. In the field, nouns often represent key entity concepts, such as "lithium-ion battery" and "manufacturing process"; verbs reflect key actions or behaviors, such as "search for". These words can be preferentially considered as keywords. At the same time, combine the stop-word filtering method. Stop words are words that frequently appear in the text but have little effect on expressing the core meaning, such as the common "of", "is", "in", "how", "about", etc. in the text. By pre-establishing a stop-word list, remove the stop words in the retrieval problem. After this series of processing, screen out the keywords corresponding to this retrieval problem.

[0026] Then, natural language processing techniques, such as word vector models, semantic role labeling, etc., are used to calculate the semantic similarity between the retrieval question and the retrieval result. For example, by converting the sentences in the question and the result into vector space representation, and then using methods such as cosine similarity to measure the semantic distance between them, the closer the distance, the higher the association strength, and the sum of the numerical values ​​of semantic similarity and frequency of occurrence is calculated and recorded as the association value of the pre-selected result, and then the pre-selected results are screened according to the discrete value to obtain the results to be analyzed, and here the pre-selected results whose association values ​​of the pre-selected results are lower than the overall discrete value are eliminated, and the remaining pre-selected results are marked as the results to be analyzed, and the specific calculation of the discrete value is a prior art, which will not be described in detail here, and at the same time, the mean of the sum of the numerical values ​​of the association values ​​of the results to be analyzed is calculated, and recorded as the association index between the model to be analyzed and the retrieval question; Similarly, the relevance indexes of all retrieval models are calculated and sorted from large to small according to the relevance index; Step 4. Next, the relevance index of each retrieval model is compared with the threshold set by the operator according to actual needs. For example, considering that this retrieval task has a high requirement for the accuracy of the retrieval results, the operator sets the threshold to 1.2. The retrieval models with a relevance index greater than the threshold are screened out and marked as pre-selected models. Assuming that among the existing five retrieval models, the relevance index of the model based on full-text semantic retrieval is 1.5, and the relevance index of the model based on the joint retrieval of classification number and keywords is 1.3, and they are both greater than the threshold of 1.2, then these two models are determined as pre-selected models. For each pre-selected model, we need to obtain its corresponding retrieval time. Retrieval time refers to the time it takes from entering the retrieval question to obtaining the retrieval results. For example, the above model based on full-text semantic retrieval takes 4 seconds to retrieve the question "Miniaturized flexible battery technology for smart wearable devices"; the model based on the joint retrieval of classification numbers and keywords takes 2 seconds to retrieve the same question. Then, based on big data analysis, we divide the search time into three levels and assign corresponding values. Based on historical data in the search field, we assume that we divide the search time into three levels: 0-2 seconds is the "fast" level, assigned a value of 3; 3-5 seconds is the "medium" level, assigned a value of 2; more than 5 seconds is the "slow" level, assigned a value of 1. The retrieval time of each pre-selected model is matched with the above-mentioned grade interval to determine its corresponding retrieval time grade and obtain the corresponding grade assignment. For example, the retrieval time of the model based on the joint retrieval of classification number and keywords is 2 seconds, which falls in the "fast" grade interval, and its grade assignment is 3; the retrieval time of the model based on full-text semantic retrieval is 4 seconds, which is in the "medium" grade interval, and its grade assignment is 2. Next, calculate the sum of the rank assignment and the correlation index of each preselected model, and record this sum as the model benefit value. For example, for the model based on the combined retrieval of classification numbers and keywords, the correlation index is 1.3 and the rank assignment is 3, so its model benefit value is 1.3 + 3 = 4.3. For the model based on full-text semantic retrieval, the correlation index is 1.5 and the rank assignment is 2, so its model benefit value is 1.5 + 2 = 3.5.

[0027] Among all the preselected models, select the preselected model with the largest model benefit value as the standard and determine it as the recommended model. In the above example, the model benefit value of the model based on the combined retrieval of classification numbers and keywords, which is 4.3, is greater than 3.5 of the model based on full-text semantic retrieval. Therefore, the model based on the combined retrieval of classification numbers and keywords is determined as the recommended model. Subsequently, obtain the retrieval problem type corresponding to the recommended model. For example, this recommended model is mainly applicable to the retrieval problem type with clear definitions of technical features and classification numbers. Combine the retrieval problem type with the recommended model to generate recommended information. Suppose the retrieval problem type is "retrieval based on specific technical features and classification numbers", and the recommended information can be expressed as "It is recommended to use the model based on the combined retrieval of classification numbers and keywords, which is applicable to the retrieval problem based on specific technical features and classification numbers". And so on, obtain the historical retrieval data in the past, comprehensively obtain the retrieval problem types corresponding to all the retrieval models in the historical data, and generate the corresponding recommended information for each retrieval problem type according to the above process. In this way, when facing different types of retrieval problems, retrieval personnel can quickly select the most suitable retrieval model based on the generated recommended information, greatly improving the efficiency and accuracy of retrieval.

[0028] Embodiment 2, please refer to Figure 2 , this application provides a retrieval model recommendation device based on a knowledge graph, including: an information acquisition module, an information analysis module, a retrieval analysis module, and a retrieval determination module, and in combination with Figure 2 it can be known that there is a one-way electrical connection between the above functional modules.

[0029] The information acquisition module is used to acquire multiple retrieval models and the corresponding knowledge graphs; The information analysis module is used to analyze the pre-acquired retrieval problems based on the knowledge graph, obtain the retrieval results corresponding to the retrieval problems, split the retrieval problems to obtain split information, and at the same time retrieve the split information with different retrieval models respectively to obtain the corresponding retrieval results. Then analyze the content matching degree between the retrieval problems and the retrieval results, and screen the retrieval results based on the content matching degree to obtain preselected results. At the same time, sort the retrieval models according to the number of preselected results to generate model sorting information; A retrieval analysis module is used to, for each of the retrieval models, perform a retrieval based on the retrieval model for the retrieval problem, obtain the retrieval result corresponding to the retrieval model, acquire model sorting information, filter the retrieval problem to obtain keywords, and calculate the occurrence frequency of the keywords. Then, calculate the semantic similarity between the retrieval problem and the retrieval result, and calculate the sum of the values of the occurrence frequency and the semantic similarity to obtain a preselected result association value. Based on discrete value analysis, obtain the result to be analyzed, and calculate the mean of the sum of the values of the result to be analyzed association value to obtain an association index; A retrieval determination module compares the association index of the retrieval model with a threshold to filter out a preselected model, determines the level assignment of the preselected model based on the retrieval time, calculates the sum of its value and the association index to obtain a model benefit value, selects the preselected model with the largest model benefit value to generate a recommended model, and generates recommended information in combination with the retrieval type of the retrieval problem.

[0030] Meanwhile, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0031] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A retrieval model recommendation method based on knowledge graph, characterized in that: The method specifically comprises the following steps: Build a knowledge graph and obtain all retrieval models and corresponding retrieval questions, and determine the question type corresponding to the retrieval question; The search question is split to obtain split information, and the split information is searched with different search models to obtain corresponding search results. Then, the content matching degree between the search question and the search results is analyzed, and the search results are screened based on the content matching degree to obtain pre-selected results. At the same time, the search models are sorted according to the number of pre-selected results to generate model sorting information. Obtain model ranking information, and filter the search questions to obtain keywords, and calculate the frequency of occurrence of keywords. Then calculate the semantic similarity between the search questions and the search results, and calculate the sum of the numerical values ​​of the frequency of occurrence and the semantic similarity to obtain the association value of the pre-selected results, and obtain the results to be analyzed based on discrete value analysis, and calculate the average value of the sum of the numerical values ​​of the association values ​​of the results to be analyzed to obtain the association index; The relevance index of the retrieval model is compared with the threshold to obtain the pre-selected model, and the rank assignment of the pre-selected model is determined based on the retrieval time. At the same time, the sum of the value of the pre-selected model and the relevance index is calculated to obtain the model benefit value. The pre-selected model with the largest model benefit value is selected to generate the recommended model, and the recommendation information is generated in combination with the retrieval type of the retrieval question.

2. The retrieval model recommendation method based on knowledge graph according to claim 1, characterized in that: The specific method of splitting the search question to obtain split information is: Obtain the search question and use the bag-of-words model to split the search question. The bag-of-words model will decompose the search question into independent words or phrases with domain significance during splitting to obtain corresponding split information.

3. The retrieval model recommendation method based on knowledge graph according to claim 1, characterized in that: The specific method of generating model sorting information is as follows: Get all the retrieval models and label them as i, where i=1, 2, ..., j, where j represents the number of retrieval models. Get the retrieval result Ci corresponding to the retrieval model i, get the split information, and calculate the proportion of the number of split information in the retrieval result Ci. And compare it with the preset value, and the search results with a screening quantity ratio greater than the preset value are recorded as pre-selected results; In this way, all pre-selected results corresponding to the retrieval model i are screened, and sorted from large to small according to the number of screened results to generate model sorting information.

4. The retrieval model recommendation method based on knowledge graph according to claim 1, characterized in that: The specific method of screening the search question to obtain keywords is: Obtain a search question, and use natural language processing tools to analyze the search question according to lexical rules, and use stop word filtering methods to comprehensively process the search question, screen out keywords corresponding to the search question, and label them as n, where n=1, 2, ..., m, where m represents the number of keywords.

5. The retrieval model recommendation method based on knowledge graph according to claim 1, characterized in that: The specific method of calculating the occurrence frequency of keywords is: The pre-selected result a is processed into sentences to obtain sentence information, and the number of occurrences of the keyword n ​​in the sentence information is calculated, and the ratio of the number of occurrences to the total number of sentence information is calculated; Similarly, the proportion values ​​corresponding to the keyword n ​​in all sentence information are calculated, and then the calculated proportion values ​​are summed up to obtain the occurrence frequency of the keyword n.

6. The retrieval model recommendation method based on knowledge graph according to claim 1, characterized in that: The specific method of calculating the average value of the sum of the correlation values ​​of the results to be analyzed to obtain the correlation index is: Using the word vector model, the semantic similarity between the search question and the search results is calculated, and the sum of the semantic similarity and the frequency of occurrence is calculated as the pre-selected result association value. Then, the pre-selected results are filtered according to the discrete value to obtain the results to be analyzed. At the same time, the mean of the sum of the association values ​​of the results to be analyzed is calculated and recorded as the association index between the model to be analyzed and the search question. The relevance indexes of all retrieval models are calculated in this way and sorted from large to small according to the relevance index.

7. The retrieval model recommendation method based on knowledge graph according to claim 1, characterized in that: The specific method of obtaining the model benefit value is: The correlation index corresponding to the retrieval model is obtained and compared with the threshold. The retrieval model with a correlation index greater than the threshold is screened and recorded as the pre-selected model. Then the retrieval time corresponding to the pre-selected model is obtained, and the retrieval time is divided into level intervals based on big data analysis, and corresponding assignment processing is performed. At the same time, the retrieval time of the pre-selected model is matched with the level interval to determine the retrieval time level corresponding to the pre-selected model, and the corresponding level assignment is obtained. Then the sum of the level assignment and the correlation index is calculated and recorded as the model benefit value.

8. The retrieval model recommendation method based on knowledge graph according to claim 1, characterized in that: The specific method of generating recommendation information in combination with the search type of the search question is: Select the pre-selected model with the largest model benefit value as the standard, generate the recommended model, obtain the corresponding search question type, and combine it with the recommended model to generate recommendation information; Similarly, historical data is obtained, and the search types corresponding to all deceleration models are obtained to generate corresponding recommendation information.

9. A retrieval model recommendation device based on a knowledge graph, used to execute the retrieval model recommendation method based on a knowledge graph according to any one of claims 1 to 8, characterized in that: include: Information acquisition module, used to obtain multiple retrieval models and corresponding knowledge graphs; An information analysis module, used to analyze the pre-acquired search questions based on the knowledge graph to obtain search results corresponding to the search questions; A retrieval analysis module, configured to perform a search for each retrieval model based on the retrieval model for the retrieval question, and obtain a retrieval result corresponding to the retrieval model; The retrieval determination module is used to determine the corresponding target retrieval model according to the retrieval results corresponding to all the retrieval models and generate recommendation information.

Images