Personalized data retrieval method and system based on artificial intelligence
Through the neural network model, the keywords and genetic algorithm sorting input by users are transformed into the neural network model, and the problems of inaccurate search results and inconsistency of personalized needs in traditional search engines are solved, and more accurate and relevant search results are achieved.
Patent Information
- Application Number
- CN202410787886.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-18
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-06-18
AI Technical Summary
When searching data, traditional search engines cannot meet users' personalized needs due to inaccurate keyword matching and single ordering. This results in inaccurate and reasonable enough search results.
A neural network model is used to convert the keywords input by users into search information of standard thesaurus, and the search results are sorted through the genetic algorithm model, taking into account data personalization and correlation.
Improve the accuracy and relevance of search results, allowing you to more accurately understand user intentions and meet personalized needs.
Smart Images

Figure CN118568355B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of artificial intelligence technology, and in particular relates to a personalized data retrieval method and system based on artificial intelligence. Background Art
[0002] Currently, when users use search engines, they are often faced with a massive amount of data search results. How to quickly and accurately find information that meets their needs from these data has become an urgent problem to be solved. Traditional search engines mostly search based on keyword matching. However, since the keywords entered by users may be ambiguous, incomplete or too broad, the search results are often not accurate enough and cannot meet the personalized needs of users.
[0003] In the prior art, although some search engines based on user portraits and personalized recommendations have emerged, traditional search engines often use simple text matching or rule-based conversion methods to convert keywords entered by users into search information. These methods cannot fully understand the user's true intentions, resulting in inaccurate conversion results. When sorting search results, the prior art mostly sorts based on single indicators such as keyword matching or web page weight, and lacks the ability to comprehensively consider factors such as data personalization and data relevance, resulting in unreasonable sorting results that cannot meet the personalized needs of users. Summary of the invention
[0004] The purpose of the present invention is to improve the personalized needs and accuracy of search results, and to obtain display information of the search results homepage in response to the first search information of the user in the search engine input box. The specific technical solution is as follows:
[0005] In a first aspect, the present invention provides a personalized data retrieval method based on artificial intelligence, the method comprising the following steps:
[0006] Step S1, based on a preset neural network model, converting first search information input by a user into a search engine input box into second search information based on a standard vocabulary.
[0007] Furthermore, the preset neural network model includes an input layer, an embedding layer, a hidden layer and an output layer.
[0008] The input layer receives first search information input by a user, and passes original text or keywords as initial data to the embedding layer.
[0009] The embedding layer converts the original text or keywords passed by the input layer into word embedding vectors. The output of the embedding layer is a vector sequence E, whose size is ,in is the dimension of the word embedding vector.
[0010] Furthermore, the hidden layer includes two sublayers. The first hidden layer is an RNN layer, which is used to receive the word embedding vector sequence E output by the embedding layer. At each time step, the RNN layer calculates a new hidden state based on the current word embedding vector and the hidden state of the previous time step. The output of the first hidden layer is H, whose size is ,That is the length of the input text, is the dimension of the hidden state.
[0011] The second hidden layer is a fully connected layer, which is used to receive the output H of the first hidden layer and convert it into a vector of the same dimension as the word segmentation vector in the standard vocabulary. The second hidden layer contains a weight matrix and a bias term. The output of the second hidden layer is C, whose size is ,in is the dimension of the word vector in the standard vocabulary, each vector in C Each corresponds to a word segmentation vector in the standard vocabulary.
[0012] The output layer aggregates the output C of the second hidden layer to generate the final second search information, which is represented by Y and has a size of , which has the same dimension as the word segmentation vector in the standard vocabulary; the function of the output layer is: , The average aggregation function.
[0013] Step S2, performing a search to obtain search results according to the user's confirmation operation on the second search information, sorting the homepage of the search results by a preset genetic algorithm model and displaying them to the user.
[0014] Furthermore, the user portrait is retrieved. On the premise that the homepage of the search results satisfies the constraint conditions, the objective function of the search results satisfies the maximization of the sum of data personalization and data relevance. The mathematical expression of the constraint conditions is:
[0015]
[0016] in, The number of results containing personalized data in the displayed page. The maximum number of results allowed on the first page of search results. When the data contains personalized data, =1, otherwise it is recorded as 0; , The number of results finally displayed on the search results home page.
[0017] The objective function of the retrieval result is:
[0018]
[0019]
[0020]
[0021] in, is the objective function of the retrieval results, is the data personalization function, is the data correlation function, and are the weight values of data personalization function and data relevance function respectively, ; Representative The data personalization index of the search results. ,when , it means that there is no correlation between the search result data and the user profile. The closer the value is to 1, the closer the search result data is to the user portrait; is the total similarity between the search results on the home page and the second search information, For the The data relevance of the search results.
[0022] Furthermore, the preset neural network model is obtained based on the following steps:
[0023] Step S21, obtaining a training set of the neural network, wherein the training set includes first search information historically input by multiple users and second search information manually determined based on a standard vocabulary.
[0024] Step S22, select a sample set, randomly select 75% of the sample set as a training set for training the neural network, and use the remaining 25% of the sample set as a verification set to verify the trained neural network.
[0025] Furthermore, the first search information of the user's historical input in the training set is input to obtain the output results of all times. , then the error value of the neural network is ,in, is the number of samples in the training set, is the similarity between the search terms output by the neural network and the standardized search terms, For the The neural network outputs the search terms for the sample set. The first input of the neural network A second search information.
[0026] When the error value is greater than a preset threshold, the time step of the neural network and the weight value of the second hidden layer are adjusted based on the deep learning strategy, and training is performed again.
[0027] In the neural network, the time steps of the first hidden layer, the second hidden layer and the output layer are the same and are all set to , the weight adjustment formula of the second hidden layer is:
[0028] , where The second hidden layer After the adjustment, the weight matrix Line The weight value of the column, For the After the adjustment, the weight matrix of the second hidden layer Line The weight value of the column, Error Partial derivatives with respect to the weights.
[0029] Step S23, when the error value is within the preset threshold range, stop training the neural network, and use the verification set to verify the accuracy of the input result of the neural network. If the accuracy is greater than 95%, stop training; otherwise, adjust the time steps of the first hidden layer, the second hidden layer, and the output layer, and retrain.
[0030] Further, Data personalization index of search results The calculation formula is:
[0031]
[0032] in, Is the first The weight or preference value of each attribute, q is the total number of attributes in the user profile; It is Search results in The similarity of the attributes with the user portrait.
[0033] No. Data relevance of search results The calculation formula is:
[0034]
[0035] in, For the Search results and The similarity between the search results. The number of results finally displayed on the search results home page.
[0036] Furthermore, sorting the homepage of the search results by a preset genetic algorithm model and displaying it to the user includes the following steps:
[0037] Step S31, select a random number from the search results. The search results are used as the initial population of the search result homepage , calculate the objective function value of the retrieval result at the initial moment .
[0038] Step S32, the initial population Perform crossover and mutation operations to generate the number Evolutionary population , .
[0039] Step S33, randomly divide each search result in population P1 into sets, each containing search results, calculate the total similarity for each set, and take the set corresponding to the maximum value The search results are used as evolution populations , computing the evolutionary population The objective function value of the corresponding retrieval result .
[0040] Step S34, the population As the initial population, repeat steps S32-S33 until the preset number of evolutions is completed. , get the corresponding set of objective function values of the search results .
[0041] Step S35, selecting the maximum value in the target function set of the search results , its corresponding The search results are the optimal solution for the search result homepage.
[0042] Furthermore, in step S31, the selected The search results are used as the initial population P0 of the search result homepage, which contains at least The search results contain personalized data. , the search result is considered to be a search result containing personalized data; after each evolution, the selected search results contain at least The search results contain personalized data.
[0043] Furthermore, the preset number of evolutions satisfy: .
[0044] Furthermore, the total similarity between the search result of the home page and the second search information The calculation formula is:
[0045]
[0046] in, For the The keyword set of the search results, It is the keyword set corresponding to the second search information.
[0047] Furthermore, the preset neural network model and the preset genetic algorithm model are updated through the user's browsing behavior on the search results; the browsing behavior includes: clicked links, dwell time, scroll depth and whether to look back, and the user's satisfaction with the search results is evaluated based on the browsing behavior.
[0048] In a second aspect, the present invention provides an artificial intelligence-based personalized data retrieval system, the system comprising a retrieval information standardization module and a retrieval result sorting module connected thereto, for executing the artificial intelligence-based personalized data retrieval method described in the first aspect.
[0049] Furthermore, the search information standardization module is used to convert the first search information input by the user into the search engine input box into the second search information based on the standard vocabulary based on a preset neural network model.
[0050] Furthermore, the search result sorting module is used to perform a search to obtain search results according to the user's confirmation operation on the second search information, sort the homepage of the search results through a preset genetic algorithm model and display them to the user.
[0051] Compared with the prior art, the present invention has the following beneficial effects:
[0052] The present invention adopts a neural network model to convert keywords input by users into retrieval information based on a standard vocabulary. Through the processing of the embedding layer and the hidden layer, it can more accurately understand the user's true intention, and use the genetic algorithm model to sort the retrieval results. Based on user portraits, by comprehensively considering factors such as data personalization and data correlation, the reasonable sorting of retrieval results is achieved, thereby improving the accuracy and relevance of the retrieval results. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 This is a flow chart of the personalized data retrieval method based on artificial intelligence of the present invention.
[0054] Figure 2 It is a schematic diagram of the composition of the personalized data retrieval system based on artificial intelligence of the present invention. DETAILED DESCRIPTION
[0055] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are 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.
[0056] Example 1
[0057] like Figure 1 FIG. 1 is a flow chart of a personalized data retrieval method based on artificial intelligence according to the present invention, and the method comprises the following steps:
[0058] Step S1, based on a preset neural network model, converting first search information input by a user into a search engine input box into second search information based on a standard vocabulary.
[0059] In this step, our main goal is to use a pre-trained neural network model to convert the first search information (which may be natural language text or keywords) entered by the user into a more standardized second search information that matches the standard vocabulary. The benefit of doing so is that it can improve the accuracy and efficiency of the search, because the words in the standard vocabulary are usually optimized and screened, and can better reflect the user's real search intent.
[0060] The preset neural network model includes an input layer, an embedding layer, a hidden layer and an output layer.
[0061] The input layer receives first search information input by a user, and passes original text or keywords as initial data to the embedding layer.
[0062] The embedding layer converts the original text or keywords passed by the input layer into word embedding vectors. Word embedding is a technique that maps words or phrases in a vocabulary to a continuous vector space of fixed size. This model uses pre-trained word embedding models (such as Word2Vec, GloVe, etc.) to obtain word embedding vectors (i.e., the number of word embedding vectors). The output of the embedding layer is a vector sequence E, whose size is ,in is the dimension of the word embedding vector.
[0063] The hidden layer includes two sub-layers, which are used to perform nonlinear transformation on the word embedding vector output by the embedding layer to capture the semantic information and context information of the user query. This model uses a recurrent neural network (RNN) or its variants (such as LSTM, GRU) as the main structure of the hidden layer.
[0064] The first hidden layer is an RNN layer, which is used to receive the word embedding vector sequence E output by the embedding layer. At each time step, the RNN layer calculates a new hidden state based on the current word embedding vector and the hidden state of the previous time step. The output of the first hidden layer is H, whose size is ,That is the length of the input text, is the dimension of the hidden state.
[0065] The second hidden layer is a fully connected layer, which is used to receive the output H of the first hidden layer and convert it into a vector of the same dimension as the word segmentation vector in the standard vocabulary. The second hidden layer contains a weight matrix and a bias term. The output of the second hidden layer is C, whose size is ,in is the dimension of the word vector in the standard vocabulary, each vector in C Each corresponds to a word segmentation vector in the standard vocabulary.
[0066] The output layer aggregates the output C of the second hidden layer to generate the final second search information, which is represented by Y and has a size of , which has the same dimension as the word segmentation vector in the standard vocabulary; the function of the output layer is: , is the average aggregation function; it should be noted that the aggregation method can be selected according to specific needs, such as average, maximum value, weighted average, etc. In this model, we use the average method, where the weight can be dynamically adjusted based on the user's historical behavior or context information.
[0067] Step S2, performing a search to obtain search results according to the user's confirmation operation on the second search information, sorting the homepage of the search results by a preset genetic algorithm model and displaying them to the user.
[0068] It should be noted that in step S1, the first search information input by the user is converted into second search information based on a standard vocabulary, and step S2 performs a search based on the second search information and processes the search results to meet the user's personalized needs.
[0069] Once the user confirms the second search information, the system starts the search process and searches the database for results related to the second search information. After the search is completed, the system obtains a list of multiple results that may be related to the user's search intent, but their sorting and display methods may not fully meet the user's personalized needs. To solve this problem, we introduced a preset genetic algorithm model to sort the homepage of search results. Genetic algorithm is an optimization algorithm that simulates the biological evolution process. It searches for the optimal solution to the problem by simulating natural selection and genetic mechanisms. In this scenario, we apply genetic algorithms to the sorting problem of search results to maximize data personalization and data relevance.
[0070] First, the system retrieves the user portrait, which is a database containing information such as user interests, preferences, and behaviors. The user portrait can help us understand the user's personalized needs and sort the search results accordingly. Then, we define the objective function and constraints. The objective function is an indicator for evaluating the quality of the search results. It consists of a data personalization function and a data relevance function, which represent the degree of match between the search results and the user portrait and the degree of relevance between the search results. The constraints limit the number of displays on the search results homepage and the amount of personalized data.
[0071] Specifically, the mathematical expression of the constraint condition ensures that the number of results containing personalized data in the displayed page is within the allowed range and does not exceed the maximum number of results allowed on the first page of the search results. The degree of match between the search results and the user profile is measured by calculating the square of the sum of the data personalization indexes of all search results; The quality of the retrieval result is evaluated by considering the total similarity between the retrieval result and the second retrieval information and the correlation between the retrieval results.
[0072] In the genetic algorithm, each possible sorting scheme is regarded as an "individual", and the fitness of the individual is evaluated by the value of the objective function. The genetic algorithm iteratively optimizes the fitness of the individual by simulating natural selection and genetic processes, and finally finds the sorting scheme that makes the objective function optimal. After optimization by the genetic algorithm, we will get a sorted search result list, in which the display content of the homepage maximizes the sum of data personalization and data relevance under the premise of meeting the constraints. Finally, the system will display this sorted list to the user to meet the user's personalized search needs.
[0073] Retrieve user portraits, which are constructed based on a large amount of user data to describe and understand the virtual representation of the target user group. In the context of search engines, user portraits are essential for providing personalized search experiences and results. Here are some key content and construction elements about user portraits: User ID: an ID that uniquely identifies each user; Name / Nickname: the user's real name or online nickname; Age / Birthday: the user's age range or specific date of birth; Gender: the user's gender; Geographic location: the user's geographical location, which can be a country, province, city, etc.; Search keywords: a list of keywords that the user has searched in the past; Search frequency: the frequency of user searches and active time periods; Click records: specific records of users clicking on search results; Interest classification: user interests are divided into different categories based on the user's search history and click behavior, such as technology, sports, entertainment, fashion, etc.; Interest strength: the strength or importance of each interest category; Preferred brands / products: a list of brands or products that the user prefers; Search habits: the user's search habits, such as whether to use long-tail keywords, whether to frequently modify search terms, etc.; Interactive behavior: the user's interactive behavior on the search results page, such as comments, sharing, likes, etc.; Visit duration: the time the user stays on the search results page.
[0074] User portrait is a dynamic data structure that needs to be continuously updated as user behavior changes. The system should regularly collect and analyze the latest user data to update the user portrait and optimize the search experience.
[0075] Under the premise that the homepage of the search results meets the constraint conditions, the objective function of the search results satisfies the maximization of the sum of data personalization and data relevance. The mathematical expression of the constraint conditions is:
[0076]
[0077] in, The number of results containing personalized data in the displayed page. The maximum number of results allowed on the first page of search results. When the data contains personalized data, =1, otherwise it is recorded as 0; , The number of results finally displayed on the search results home page.
[0078] The objective function of the retrieval result is:
[0079]
[0080]
[0081]
[0082] in, is the objective function of the retrieval results, is the data personalization function, is the data correlation function, and are the weight values of data personalization function and data relevance function respectively, ; Representative The data personalization index of the search results. ,when , it means that there is no correlation between the search result data and the user profile. The closer the value is to 1, the closer the search result data is to the user portrait; is the total similarity between the search results on the home page and the second search information, For the The data relevance of the search results.
[0083] The preset neural network model is obtained based on the following steps:
[0084] Step S21, obtaining a training set of the neural network, wherein the training set includes first search information historically input by multiple users and second search information manually determined based on a standard vocabulary.
[0085] Step S22, select a sample set, randomly select 75% of the sample set as a training set for training the neural network, and use the remaining 25% of the sample set as a validation set to validate the trained neural network.
[0086] Input the first search information of the user's historical input in the training set and get the previous output results , then the error value of the neural network is ,in, is the number of samples in the training set, is the similarity between the search terms output by the neural network and the standardized search terms, For the The neural network outputs the search terms for the sample set. The first input of the neural network A second search information.
[0087] When the error value is greater than a preset threshold, the time step of the neural network and the weight value of the second hidden layer are adjusted based on the deep learning strategy, and training is performed again.
[0088] In the neural network, the time steps of the first hidden layer, the second hidden layer and the output layer are the same and are all set to , the weight adjustment formula of the second hidden layer is:
[0089] , where The second hidden layer After the adjustment, the weight matrix Line The weight value of the column, For the After the adjustment, the weight matrix of the second hidden layer Line The weight value of the column, Error Partial derivatives with respect to the weights.
[0090] Step S23, when the error value is within the preset threshold range, stop training the neural network, and use the verification set to verify the accuracy of the input result of the neural network. If the accuracy is greater than 95%, stop training; otherwise, adjust the time steps of the first hidden layer, the second hidden layer, and the output layer, and retrain.
[0091] No. Data personalization index of search results The calculation formula is:
[0092]
[0093] in, Is the first The weight or preference value of each attribute, q is the total number of attributes in the user profile; It is Search results in The similarity of the attributes with the user portrait.
[0094] No. Data relevance of search results The calculation formula is:
[0095]
[0096] in, For the Search results and The similarity between the search results. The number of results finally displayed on the search results home page.
[0097] Sorting the homepage of search results through a preset genetic algorithm model and displaying it to the user includes the following steps:
[0098] Step S31, select a random number from the search results. The search results are used as the initial population of the search result homepage , calculate the objective function value of the retrieval result at the initial moment .
[0099] Step S32, the initial population Perform crossover and mutation operations to generate the number Evolutionary population , .
[0100] Step S33, randomly divide each search result in population P1 into sets, each containing search results, calculate the total similarity for each set, and take the set corresponding to the maximum value The search results are used as evolution populations , computing the evolutionary population The objective function value of the corresponding retrieval result .
[0101] Step S34, the population As the initial population, repeat steps S32-S33 until the preset number of evolutions is completed. , get the corresponding set of objective function values of the search results .
[0102] Step S35, selecting the maximum value in the target function set of the search results , its corresponding The search results are the optimal solution for the search result homepage.
[0103] In step S31, the selected The search results are used as the initial population P0 of the search result homepage, which contains at least The search results contain personalized data. , the search result is considered to be a search result containing personalized data; after each evolution, the selected search results contain at least The search results contain personalized data.
[0104] The preset number of evolutions satisfy: .
[0105] The total similarity between the search result of the home page and the second search information The calculation formula is:
[0106]
[0107] in, For the The keyword set of the search results, It is the keyword set corresponding to the second search information.
[0108] The preset neural network model and the preset genetic algorithm model are updated through the user's browsing behavior on the search results; the browsing behavior includes: clicked links, dwell time, scroll depth and whether to look back, and the user's satisfaction with the search results is evaluated based on the browsing behavior.
[0109] Based on the above browsing behavior, a satisfaction evaluation model can be constructed, which converts the user's browsing behavior into a satisfaction score to quantify the user's satisfaction with the retrieval results; for example, clicking a link can give a higher score, while a short stay and replay behavior may lower the score.
[0110] To update the neural network model, we need to collect the user's browsing behavior data and the corresponding search results. These data will be used to train and optimize the neural network model. Using the collected data, we can train the neural network model to predict the user's satisfaction with the search results. The goal of the model is to learn the mapping relationship from user browsing behavior to satisfaction. As new data is continuously collected, we can regularly update the neural network model to adapt to changes in user behavior and the development of search needs. Through iterative training and optimization, the model can gradually improve its prediction accuracy and generalization ability.
[0111] Updating the genetic algorithm model requires the use of a fitness function to evaluate the pros and cons of individuals. The fitness function can be adjusted according to the user satisfaction score to make the sorted search results more in line with the user's personalized needs. By analyzing the user's browsing behavior, we can find some parameters that have an important impact on the sorting of search results (such as the weight values of the data personalization function and the data relevance function). Genetic algorithms are used to optimize these parameters to find the optimal solution that maximizes user satisfaction. As user satisfaction continues to improve and new data is continuously collected, we can regularly iterate the genetic algorithm model to further improve the personalization and relevance of search results.
[0112] Example 2
[0113] like Figure 2 As shown, this embodiment provides a personalized data retrieval system based on artificial intelligence, which includes a retrieval information standardization module and a retrieval result sorting module connected in sequence, and is used to execute the personalized data retrieval method based on artificial intelligence described in Example 1.
[0114] The search information standardization module is used to convert the first search information input by the user into the search engine input box into the second search information based on the standard word library based on the preset neural network model.
[0115] The search result sorting module is used to perform a search to obtain search results according to the user's confirmation operation on the second search information, sort the homepage of the search results through a preset genetic algorithm model and display them to the user.
[0116] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. Personalized data retrieval method based on artificial intelligence, It is characterized in that The method responds to the first search information input by the user into the search engine box, and the method comprises the following steps: Step S1, converting first search information input into a search engine input box by a user into second search information based on a standard word library based on a preset neural network model; The preset neural network model includes an input layer, an embedding layer, a hidden layer and an output layer; The input layer receives the first search information input by the user and passes the original text or keywords as initial data to the embedding layer; The embedding layer converts the original text or keywords passed by the input layer into word embedding vectors. The output of the embedding layer is a vector sequence E, whose size is ,in is the dimension of the word embedding vector; The hidden layer includes two sublayers. The first hidden layer is an RNN layer, which is used to receive the word embedding vector sequence E output by the embedding layer. At each time step, the RNN layer calculates a new hidden state based on the current word embedding vector and the hidden state of the previous time step. The output of the first hidden layer is H, whose size is ,That is the length of the input text, is the dimension of the hidden state; The second hidden layer is a fully connected layer, which is used to receive the output H of the first hidden layer and convert it into a vector of the same dimension as the word segmentation vector in the standard vocabulary. The second hidden layer contains a weight matrix and a bias term. The output of the second hidden layer is C, whose size is ,in is the dimension of the word vector in the standard vocabulary, each vector in C Each corresponds to a word segmentation vector in the standard vocabulary; The output layer aggregates the output C of the second hidden layer to generate the final second search information, which is represented by Y and has a size of , which has the same dimension as the word segmentation vector in the standard vocabulary; the function of the output layer is: , is the average aggregation function; Step S2, performing a search to obtain search results according to the user's confirmation operation on the second search information, sorting the homepage of the search results by a preset genetic algorithm model and displaying the sorted search results to the user; Retrieve the user portrait. On the premise that the homepage of the search results meets the constraint conditions, the objective function of the search results satisfies the maximization of the sum of data personalization and data relevance. The mathematical expression of the constraint conditions is: in, The number of results containing personalized data in the displayed page. The maximum number of results allowed on the first page of search results. When the data contains personalized data, =1, otherwise it is recorded as 0; , The number of results finally displayed on the search result homepage; The objective function of the retrieval result is: in, is the objective function of the retrieval results, is the data personalization function, is the data correlation function, and are the weight values of data personalization function and data relevance function respectively, ; Representative The data personalization index of the search results. ,when , it means that there is no correlation between the search result data and the user profile. The closer the value is to 1, the closer the search result data is to the user portrait; is the total similarity between the search results on the home page and the second search information, For the The data relevance of the search results.
2. According to the personalized data retrieval method based on artificial intelligence according to claim 1, It is characterized in that The preset neural network model is obtained based on the following steps: Step S21, obtaining a training set of the neural network, wherein the training set includes first search information historically input by multiple users and second search information manually determined based on a standard word library; Step S22, selecting a sample set, randomly selecting 75% of the sample set as a training set for training the neural network, and using the remaining 25% of the sample set as a validation set to validate the trained neural network; Input the first search information of the user's historical input in the training set and get the previous output results , then the error value of the neural network is ,in, is the number of samples in the training set, is the similarity between the search terms output by the neural network and the standardized search terms, For the The neural network outputs the search terms for the sample set. The first input of the neural network Second search information; When the error value is greater than a preset threshold, the time step of the neural network and the weight value of the second hidden layer are adjusted based on the deep learning strategy, and training is performed again; In the neural network, the time steps of the first hidden layer, the second hidden layer and the output layer are the same and are all set to , the weight adjustment formula of the second hidden layer is: , where The second hidden layer After the adjustment, the weight matrix Line The weight value of the column, For the After the adjustment, the weight matrix of the second hidden layer Line The weight value of the column, Error Partial derivatives with respect to weight values; Step S23, when the error value is within the preset threshold range, stop training the neural network, and use the verification set to verify the accuracy of the input result of the neural network. If the accuracy is greater than 95%, stop training; otherwise, adjust the time steps of the first hidden layer, the second hidden layer, and the output layer, and retrain.
3. According to the personalized data retrieval method based on artificial intelligence according to claim 2, It is characterized in that No. Data personalization index of search results The calculation formula is: in, Is the first The weight or preference value of each attribute, q is the total number of attributes in the user profile; It is Search results in Similarity with the user portrait in terms of attributes; No. Data relevance of search results The calculation formula is: in, For the Search results and The similarity between the search results. The number of results finally displayed on the search results home page.
4. According to the personalized data retrieval method based on artificial intelligence according to claim 3, It is characterized in that Sorting the homepage of search results through a preset genetic algorithm model and displaying it to the user includes the following steps: Step S31, select a random number from the search results. The search results are used as the initial population of the search result homepage , calculate the objective function value of the retrieval result at the initial moment ; Step S32, the initial population Perform crossover and mutation operations to generate the number Evolutionary population , ; Step S33, randomly divide each search result in population P1 into sets, each containing search results, calculate the total similarity for each set, and take the set corresponding to the maximum value The search results are used as evolution populations , computing the evolving population The objective function value of the corresponding retrieval result ; Step S34, the population As the initial population, repeat steps S32-S33 until the preset number of evolutions is completed. , get the corresponding set of objective function values of the search results ; Step S35, selecting the maximum value in the target function set of the search results , its corresponding The search results are the optimal solution for the search result homepage.
5. According to the personalized data retrieval method based on artificial intelligence according to claim 4, It is characterized in that In step S31, the selected The search results are used as the initial population P0 of the search result homepage, which contains at least The search results contain personalized data. , the search result is considered to be a search result containing personalized data; after each evolution, the selected search results contain at least The search results contain personalized data.
6. According to the personalized data retrieval method based on artificial intelligence according to claim 5, It is characterized in that The preset number of evolutions satisfy: .
7. According to the personalized data retrieval method based on artificial intelligence according to claim 5, It is characterized in that The total similarity between the search result of the home page and the second search information The calculation formula is: in, For the The keyword set of the search results, It is the keyword set corresponding to the second search information.
8. According to the personalized data retrieval method based on artificial intelligence according to claim 7, It is characterized in that The preset neural network model and the preset genetic algorithm model are updated through the user's browsing behavior on the search results; the browsing behavior includes: clicked links, dwell time, scroll depth and whether to look back, and the user's satisfaction with the search results is evaluated based on the browsing behavior.
9. Personalized data retrieval system based on artificial intelligence, It is characterized in that The system includes a retrieval information standardization module and a retrieval result sorting module connected thereto, for executing the personalized data retrieval method based on artificial intelligence according to any one of claims 1 to 8; The search information standardization module is used to convert the first search information input into the search engine input box by the user into the second search information based on the standard word library based on the preset neural network model; The search result sorting module is used to perform a search to obtain search results according to the user's confirmation operation on the second search information, sort the homepage of the search results through a preset genetic algorithm model and display them to the user.
Citation Information
Patent Citations
Artificial intelligence Internet of Things data information measurement, transmission and analysis method
CN114793304A
Information retrieval method fusing deep belief neural network and whale Manhattan optimization algorithm
CN117688218A
Internet mass data accurate search method and system based on AI technology
CN118093982A
Expansion of Search Queries Using Information Categorization
US20100145975A1
