User behavior big data analysis method and system based on artificial intelligence

Abstract

The present invention relates to the field of artificial intelligence technology, and relates to a method and system for analyzing user behavior big data based on artificial intelligence. The present invention can capture the user's browsing behavior and points of interest more comprehensively by acquiring the user's browsing behavior big data and determining the multi-source session data of multiple knowledge point labels associated therewith. Furthermore, by generating derived multi-source session data, the feature expression data is effectively enriched, and the accuracy and depth of the analysis are improved. By extracting the session coding vector from the derived multi-source session data and combining it with the browsing path graph vector of the user's browsing behavior big data, accurate prediction of user preferences is achieved, and corresponding preference label data is generated, which not only improves the accuracy of recommendations, but also provides users with a more personalized selection experience, thereby optimizing the user's browsing and selection process, and enhancing user satisfaction and platform usage efficiency.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based user behavior big data analysis method and system. Background Art

[0002] In the current Internet market, user behavior analysis is crucial to improving user experience, optimizing the house recommendation system, and improving platform operation efficiency. Traditional user behavior analysis methods often rely on simple browsing history statistics and click-through rate analysis. These methods are incapable of processing large-scale, multi-dimensional user behavior data and are difficult to accurately capture users' deep preferences and needs.

[0003] With the rapid development of artificial intelligence technology, especially the continuous maturity of deep learning, natural language processing and other technologies, new ideas and methods have been provided for user behavior analysis. With artificial intelligence technology, user behavior data can be deeply mined and analyzed, so as to more accurately understand the intentions and preferences behind user behavior. However, the existing user behavior analysis methods based on artificial intelligence still have some limitations when processing house browsing behavior data. For example, these methods often only focus on users' direct clicks or browsing behaviors, ignoring the indirect multi-source session data generated by users during the browsing process. Summary of the invention

[0004] In view of this, the purpose of this application is to provide a user behavior big data analysis method and system based on artificial intelligence.

[0005] According to a first aspect of the present application, there is provided a method for analyzing user behavior big data based on artificial intelligence, which is applied to a housing recommendation system, and the method comprises:

[0006] Obtaining user house browsing behavior big data, and determining multi-source session data of multiple house knowledge point tags associated with the user house browsing behavior big data;

[0007] Determine a multi-source conversation knowledge vector sequence of the multiple housing source knowledge point labels according to the multi-source conversation data of the multiple housing source knowledge point labels;

[0008] Based on the multi-source conversation knowledge vector sequences of the multiple house knowledge point labels, generating derived multi-source conversation data of the user's house browsing behavior big data, wherein the derived multi-source conversation data includes more feature expression data than the multi-source conversation data of the multiple house knowledge point labels;

[0009] Extracting a session coding vector from the derived multi-source session data to generate a derived session coding vector;

[0010] Extracting a browsing path graph vector from the user's house browsing behavior big data;

[0011] Based on the derived session coding vector and the browsing path graph vector, the housing preference prediction is performed on the user housing browsing behavior big data, and the housing preference label data corresponding to the user housing browsing behavior big data is generated.

[0012] In a possible implementation of the first aspect, the multi-source session data of the multiple property knowledge point labels include at least two of the multi-source session data of property feature identification labels, the multi-source session data of user behavior description labels, or the multi-source session data of market trend analysis labels; wherein, the multi-source session data of the property feature identification labels are session data of property features in the user property browsing behavior big data, the multi-source session data of the user behavior description labels are descriptive session data of user browsing behavior patterns in the user property browsing behavior big data, and the multi-source session data of the market trend analysis labels are session data of target market trend knowledge points in the user property browsing behavior big data.

[0013] In a possible implementation of the first aspect, determining multi-source session data of multiple housing knowledge point tags associated with the user housing browsing behavior big data includes:

[0014] Identify the user's house browsing behavior big data, generate target market trend knowledge points in the user's house browsing behavior big data and association information of the target market trend knowledge points in the user's house browsing behavior big data;

[0015] Acquiring reference knowledge data associated with the target market trend knowledge point from a set knowledge base, wherein the set knowledge base includes reference knowledge data for a plurality of market trend knowledge points;

[0016] Obtaining restriction information corresponding to the user's house browsing behavior big data, wherein the restriction information is used to limit the amount of session data describing the user's browsing behavior pattern in the user's house browsing behavior big data;

[0017] According to the reference knowledge data associated with the target market trend knowledge point and the associated information, based on the restriction information, session data describing the user browsing behavior pattern in the user housing browsing behavior big data is generated.

[0018] In a possible implementation of the first aspect, determining multi-source session data of multiple housing knowledge point tags associated with the user housing browsing behavior big data includes:

[0019] Performing behavior pattern recognition on the user's house browsing behavior big data to generate concerned house data, core browsing data and auxiliary browsing data in the user's house browsing behavior big data;

[0020] Extracting a focused property vector from the focused property data, extracting a core browsing vector from the core browsing data, and extracting an auxiliary browsing vector from the auxiliary browsing data;

[0021] Based on the concerned house source vector, the core browsing vector and the auxiliary browsing vector, session data describing the user browsing behavior pattern in the user house source browsing behavior big data is generated.

[0022] In a possible implementation of the first aspect, the generating of the derived multi-source session data of the user's housing browsing behavior big data based on the multi-source session knowledge vector sequence of the multiple housing knowledge point tags further includes:

[0023] Integrate the multi-source conversation knowledge vector sequence of the multiple house knowledge point labels to generate a multi-source conversation knowledge vector integration sequence;

[0024] Processing the multi-source conversation knowledge vector integration sequence using a self-attention mechanism to determine a self-attention coefficient of the multi-source conversation knowledge vector integration sequence;

[0025] The knowledge vector value of the multi-source conversation knowledge vector integration sequence is updated according to the self-attention coefficient of the multi-source conversation knowledge vector integration sequence, and the derived multi-source conversation data of the user's house browsing behavior big data is generated according to the updated knowledge vector value.

[0026] In a possible implementation of the first aspect, the browsing path graph vector includes a browsing path node vector of the user's house browsing behavior big data; and extracting the browsing path graph vector from the user's house browsing behavior big data includes:

[0027] Decomposing the user's house browsing behavior big data into multiple house browsing behavior paths;

[0028] Constructing a behavior pattern association graph according to the plurality of house browsing behavior paths, the behavior pattern association graph comprising a plurality of graph nodes and a plurality of association links, the plurality of graph nodes being used to reflect different house browsing behavior paths, and the plurality of association links being used to reflect behavior pattern association information between the respective house browsing behavior paths;

[0029] Graph convolution processing is performed on the behavior pattern association graph to generate a browsing path node vector corresponding to the behavior pattern association graph.

[0030] In a possible implementation of the first aspect, constructing a behavior pattern association graph according to the multiple house browsing behavior paths includes:

[0031] Respectively identifying target behavior pattern features in each of the house browsing behavior paths;

[0032] A behavior pattern association graph is constructed based on the target behavior pattern features in the multiple house browsing behavior paths, the multiple graph nodes of the behavior pattern association graph are used to reflect the target behavior pattern features in different house browsing behavior paths, and the multiple association links are used to reflect the behavior pattern association information between each of the target behavior pattern features.

[0033] In a possible implementation of the first aspect, the browsing path graph vector further includes a housing location association vector of the user housing browsing behavior big data, and the method further includes:

[0034] Constructing a location association graph according to the plurality of house browsing behavior paths, the location association graph comprising a plurality of graph nodes and a plurality of association links, the plurality of graph nodes being used to reflect different house browsing behavior paths, and the plurality of association links being used to reflect the house location association information between the house browsing behavior paths;

[0035] Perform graph convolution processing on the location association graph to generate a house location association vector corresponding to the location association graph.

[0036] In a possible implementation of the first aspect, performing housing preference prediction on the user housing browsing behavior big data based on the derived session encoding vector and the browsing path graph vector, and generating housing preference label data corresponding to the user housing browsing behavior big data, includes:

[0037] Merging the derived session encoding vector and the browsing path graph vector to generate fused vector data;

[0038] The fused vector data is input into a pre-trained housing preference prediction model to perform housing preference prediction, and generate housing preference label data corresponding to the user housing browsing behavior big data.

[0039] According to a second aspect of the present application, a housing recommendation system is provided, the housing recommendation system comprising a processor and a readable storage medium, the readable storage medium storing a program, which, when executed by the processor, implements the aforementioned artificial intelligence-based user behavior big data analysis method.

[0040] According to the third aspect of the present application, a computer-readable storage medium is provided, in which computer-executable instructions are stored. When the computer-executable instructions are monitored to be executed, the aforementioned user behavior big data analysis method based on artificial intelligence is implemented.

[0041] According to any of the above aspects, the embodiment of the present application realizes in-depth analysis and mining of the big data of user's house browsing behavior. By obtaining the big data of user's house browsing behavior and determining the multi-source session data of multiple house knowledge point labels associated therewith, the user's browsing behavior and points of interest can be captured more comprehensively. Furthermore, by generating derived multi-source session data, the feature expression data is effectively enriched, and the accuracy and depth of the analysis are improved. By extracting the session coding vector from the derived multi-source session data and combining it with the browsing path graph vector of the user's house browsing behavior big data, the accurate prediction of the user's house preference is achieved, and the corresponding house preference label data is generated, which not only improves the accuracy of house recommendation, but also provides users with a more personalized and demand-oriented house selection experience, thereby optimizing the user's browsing and selection process, and enhancing user satisfaction and platform usage efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 A schematic diagram of the process of the user behavior big data analysis method based on artificial intelligence provided in an embodiment of the present application is shown;

[0044] Figure 2 A schematic diagram of the component structure of a housing recommendation system for implementing the above-mentioned artificial intelligence-based user behavior big data analysis method provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0046] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0047] Figure 1 The flowchart of the method for analyzing user behavior big data based on artificial intelligence provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps in the method for analyzing user behavior big data based on artificial intelligence can be interchanged according to actual needs, or some steps can be omitted or deleted. The detailed steps of the method for analyzing user behavior big data based on artificial intelligence are introduced as follows.

[0048] Step S110, obtaining user house browsing behavior big data, and determining multi-source session data of multiple house knowledge point tags associated with the user house browsing behavior big data.

[0049] In this embodiment, the server receives big data of the house browsing behavior of a user named Mr. Li. In the past month, Mr. Li frequently browsed house information in different regions, including viewing detailed descriptions of houses, comparing prices of different houses, and paying attention to supporting facilities around houses.

[0050] The server first analyzes and processes the big data of these users' house browsing behavior, and determines the multi-source session data of multiple house knowledge point tags associated with these users' house browsing behavior big data. For example, the server recognizes that some of the houses browsed by Mr. Li are high-end apartments located in the city center, and the characteristics of these houses include fine decoration, high-end property management, large area, etc. Therefore, the server determines "high-end apartment characteristics in the city center" as a house knowledge point tag and obtains the multi-source session data related to it, such as other users' evaluation of high-end apartments in the city center, experts' analysis of such apartments, etc.

[0051] At the same time, the server also found that when Mr. Li browsed houses, he often compared the price fluctuations of different houses and paid attention to the changes in market supply and demand. Therefore, "market trend analysis" was also determined as a house knowledge point tag, and the server obtained relevant multi-source session data, such as recent real estate market price trend reports and policy impact analysis on housing prices.

[0052] In addition, the server analyzed Mr. Li's browsing behavior patterns and found that every time he browsed a house, he would first check the house's floor plan and then pay attention to the green environment of the community. Therefore, "user behavior description" was determined as a label, and the related multi-source session data may include other users with similar behavior patterns. The decision-making factors for buying a house.

[0053] Step S120: determining a multi-source conversation knowledge vector sequence of the multiple housing knowledge point tags according to the multi-source conversation data of the multiple housing knowledge point tags.

[0054] Take the three housing source knowledge point tags of “Characteristics of high-end apartments in the city center”, “Market trend analysis” and “User behavior description” determined in step S110 as an example.

[0055] For the multi-source conversation data of the label "Features of high-end apartments in the city center", the server may convert it into a series of vector representations. For example, "fine decoration" is represented as a specific vector [0.8, 0.2, 0.1, 0.05], where each value represents the quantitative value of the feature in different dimensions; "high-end property management" is represented as [0.7, 0.3, 0.08, 0.02]; "large area" is represented as [0.6, 0.4, 0.05, 0.01], etc. Arranging these vectors in a certain order forms a multi-source conversation knowledge vector sequence of the label "Features of high-end apartments in the city center".

[0056] For the multi-source session data of the "market trend analysis" label, assuming that the information obtained is "housing prices have been on the rise recently, and policy regulation has been strengthened", the server can represent "housing price increase" as [0.9, 0.1, 0.05, 0.01] and "policy regulation has been strengthened" as [0.8, 0.2, 0.1, 0.08], thereby forming a vector sequence for the label.

[0057] For the "user behavior description" label, such as "first check the floor plan, then pay attention to the green environment of the community", the server can represent "check the floor plan" as [0.7, 0.3, 0.1, 0.05] and "pay attention to the green environment of the community" as [0.6, 0.4, 0.08, 0.02], thereby obtaining the corresponding vector sequence.

[0058] Step S130, based on the multi-source conversation knowledge vector sequence of the multiple property knowledge point labels, generate derived multi-source conversation data of the user's property browsing behavior big data, the feature expression data contained in the derived multi-source conversation data is more than the feature expression data contained in the multi-source conversation data of the multiple property knowledge point labels.

[0059] In this embodiment, the server has obtained the multi-source conversation knowledge vector sequence of the above three house source knowledge point labels. Next, the server generates derived multi-source conversation data through a series of calculations and analyses.

[0060] For example, the server can perform weighted summation on these vector sequences. Assume that the weight of the vector sequence "Features of high-end apartments in the city center" is 0.4, the weight of the vector sequence "Market trend analysis" is 0.3, and the weight of the vector sequence "User behavior description" is 0.3. Through the weighted summation operation, a new comprehensive vector sequence is obtained.

[0061] Then, the server further expands and enriches this comprehensive vector sequence. For example, it combines Mr. Li's personal information (such as age, occupation, income, etc.) and historical house purchase records to generate more feature expressions. Assuming that Mr. Li is a young, high-paid white-collar worker who likes a modern lifestyle, the server can convert this information into a new vector and merge it with the previous comprehensive vector sequence to obtain derived multi-source session data containing more feature expression data.

[0062] Step S140, extracting a session coding vector from the derived multi-source session data, generating a derived session coding vector, and extracting a browsing path graph vector from the user's house browsing behavior big data.

[0063] The server processes the derived multi-source conversation data using encoding techniques in deep learning algorithms, for example, using a long short-term memory network (LSTM) to encode the derived multi-source conversation data.

[0064] Assuming that the derived multi-source conversation data is a series of text descriptions, the server inputs these texts into the LSTM network. The LSTM network learns its inherent patterns and features based on the input text sequence and outputs a fixed-length conversation encoding vector. This vector can effectively capture the key information in the derived multi-source conversation data.

[0065] At the same time, the server extracts the browsing path graph vector from Mr. Li's user house browsing behavior big data. First, the server decomposes Mr. Li's browsing behavior into multiple house browsing behavior paths, such as "first browsed the house in community A in the city center, then checked the house in community B in the suburbs, and finally focused on the house in community C in the city center."

[0066] Next, the server constructs a behavior pattern association graph based on these browsing behavior paths. In this graph, the browsing behavior path of each community's listings is used as a graph node, and the association links between the nodes represent Mr. Li's browsing order and attention transfer between different listings. Then, the server performs graph convolution processing on this behavior pattern association graph to generate a browsing path node vector that reflects Mr. Li's browsing behavior pattern.

[0067] Step S150, based on the derived session coding vector and the browsing path graph vector, predict the housing preference of the user's housing browsing behavior big data, and generate housing preference label data corresponding to the user's housing browsing behavior big data.

[0068] In this embodiment, the server has obtained Mr. Li's derived session coding vector and browsing path graph vector. First, the server merges the two vectors. For example, through a simple splicing operation, the derived session coding vector and the browsing path graph vector are connected into a longer vector.

[0069] The server then inputs the fused vector data into the pre-trained housing preference prediction model. This housing preference prediction model has been trained with a large amount of user data and can learn the complex relationship between different users' browsing behaviors and housing preferences. The housing preference prediction model analyzes and calculates the input fusion vector and finally predicts Mr. Li's housing preference label data. Assume that the prediction result shows that Mr. Li prefers high-end apartments in the city center, well-decorated, large area, and relatively stable prices. The server generates corresponding label data based on these preference information to provide Mr. Li with more accurate housing recommendations and services.

[0070] In this way, valuable information can be mined from the big data of users' house browsing behavior, users' house preferences can be predicted, and better services and experiences can be provided to users.

[0071] In another example, let us assume that another user, Ms. Wang, frequently browses listings within a week. After the server obtains her browsing behavior big data, it first determines the relevant multi-source session data according to step S110.

[0072] Ms. Wang mainly browses school district housing, and the tags of housing knowledge points she is interested in include "school district housing policy", "school quality assessment" and "user's description of school district housing needs". The server obtains multi-source session data related to these tags, such as the latest school district division policy documents, rankings and evaluations of different schools, and experience sharing of other parents in purchasing school district housing.

[0073] In step S120, the server converts these multi-source session data into a sequence of knowledge vectors. For example, "school district housing policy" may be represented as [0.7, 0.2, 0.1, 0.05], "school quality evaluation" may be represented as [0.8, 0.1, 0.08, 0.02], and "user demand description for school district housing" may be represented as [0.6, 0.3, 0.07, 0.01].

[0074] Then, in step S130, the server generates derived multi-source conversation data based on these vector sequences. For example, considering that Ms. Wang is a parent with two children, the server combines this information and adds features such as the emphasis on the family's living space and the richness of surrounding educational resources to generate richer derived conversation data.

[0075] In step S140, when extracting the session coding vector from the derived multi-source session data, a convolutional neural network (CNN) is used for encoding to generate a coding vector that can accurately reflect Ms. Wang's focus on school district housing. At the same time, Ms. Wang's browsing path is analyzed, a browsing path graph is constructed, and a browsing path graph vector reflecting her browsing order and focus shift is extracted.

[0076] Finally, in step S150, these two vectors are fused and input into the housing preference prediction model, which predicts that Ms. Wang prefers school district housing that is located near high-quality schools, has a large house area, and has complete surrounding supporting facilities, and generates corresponding housing preference label data.

[0077] Let’s look at another example. Mr. Zhang is a retiree. His house browsing behavior is mainly concentrated on suburban houses with beautiful environment and convenient transportation.

[0078] In step S110, the housing knowledge point tags determined by the server include "suburban environment advantages", "transportation convenience analysis" and "retirees' housing needs", and the multi-source session data obtained covers suburban air quality reports, traffic planning documents and other retirees' housing options.

[0079] Step S120 converts these multi-source conversation data into vector sequences, such as "suburban environment advantages" is represented by [0.6, 0.3, 0.08, 0.02], "traffic convenience analysis" is represented by [0.7, 0.2, 0.07, 0.01], and "retirees' housing needs" is represented by [0.5, 0.4, 0.06, 0.01].

[0080] In step S130 , considering that Mr. Zhang is very interested in medical facilities, the server incorporates relevant features of surrounding medical resources on the basis of the existing vector sequence to generate derived multi-source session data.

[0081] In step S140, the session encoding vector is extracted through a recurrent neural network (RNN), and a path graph reflecting Mr. Zhang's browsing behavior is constructed and processed to obtain a browsing path graph vector.

[0082] In step S150, the two vectors are fused and input into the model to predict that Mr. Zhang prefers suburban housing with good greenery, convenient transportation, and good medical facilities, and generate corresponding preference label data.

[0083] Suppose there is another user, Mr. Zhao, who is a young entrepreneur and is interested in small apartments in the city center.

[0084] In step S110, the housing knowledge point tags determined by the server include "characteristics of small apartments in the city center", "advantages of housing around entrepreneurial parks" and "living preferences of young entrepreneurs", and the multi-source session data obtained include layout design cases of small apartments, development plans of entrepreneurial parks, and surveys on living habits of young entrepreneurs.

[0085] In step S120, these multi-source conversation data are converted into vector sequences, for example, "characteristics of small apartments in the city center" is [0.7, 0.2, 0.08, 0.02], "advantages of housing resources around the entrepreneurial park" is [0.8, 0.1, 0.07, 0.01], and "living preferences of young entrepreneurs" is [0.6, 0.3, 0.06, 0.01].

[0086] In step S130 , the server generates derivative multi-source conversation data including these features, taking into account Mr. Zhao's needs for social space and office convenience.

[0087] Step S140, using an autoencoder to encode the derived multi-source conversation data, extracting the conversation encoding vector, and constructing and processing Mr. Zhao's browsing path graph to obtain a browsing path graph vector.

[0088] Step S150, after fusing the two vectors, input the model to predict that Mr. Zhao prefers a small-sized apartment in the city center that is located near the entrepreneurial park and has a shared office space, and generates corresponding preference label data.

[0089] Suppose there is a user, Ms. Sun, who is a homebuyer with improvement needs and is interested in large-sized improvement properties.

[0090] In step S110, the housing source knowledge point tags determined by the server include "large-sized apartment space layout", "improved housing source supporting facilities" and "user improvement demand characteristics", and the multi-source conversation data obtained include large-sized apartment model room display materials, improved community supporting service introductions and other improvement buyers' voices.

[0091] In step S120, these multi-source conversation data are converted into vector sequences, such as "large apartment space layout" is [0.7, 0.2, 0.08, 0.02], "improved housing facilities" is [0.8, 0.1, 0.07, 0.01], and "user improvement demand characteristics" is [0.6, 0.3, 0.06, 0.01].

[0092] Step S130 , the server generates richer derivative multi-source conversation data based on Ms. Sun’s high requirements for the quality of community property services and surrounding educational resources.

[0093] Step S140, using the deep belief network to extract the session coding vector, constructing and processing Ms. Sun's browsing path graph, and obtaining a browsing path graph vector.

[0094] Step S150, the two vectors are fused and input into the model to predict that Ms. Sun prefers large-sized improved housing with open space, complete supporting facilities, high-quality property services and high-quality schools nearby, and generate corresponding preference label data.

[0095] Suppose a user, Mr. Wu, is an investment home buyer who is concerned about the property's appreciation potential and return on investment.

[0096] In step S110, the house source knowledge point tags determined by the server include "real estate investment trends", "regional development potential" and "investment return analysis", and the multi-source session data obtained include investment reports on the real estate market, urban planning documents, and expert calculation cases of investment return, etc.

[0097] Step S120 , converting these multi-source conversation data into vector sequences, for example, “real estate investment trend” is [0.7, 0.2, 0.08, 0.02], “regional development potential” is [0.8, 0.1, 0.07, 0.01], and “investment return analysis” is [0.6, 0.3, 0.06, 0.01].

[0098] In step S130, the server considers Mr. Wu's risk tolerance and investment cycle expectations and generates derivative multi-source session data including these factors.

[0099] Step S140, using a generative adversarial network to extract a session coding vector, constructing and processing Mr. Wu's browsing path graph, and obtaining a browsing path graph vector.

[0100] Step S150, the two vectors are fused and input into the model to predict that Mr. Wu prefers properties located in emerging development areas, with large appreciation potential and high investment returns in the short term, and generate corresponding preference label data.

[0101] It is worth noting that the above is only a part of the embodiments of the present application, and more common examples can be expanded for those skilled in the art.

[0102] Based on the above steps, the embodiment of the present application realizes the in-depth analysis and mining of the big data of the user's house browsing behavior. By obtaining the big data of the user's house browsing behavior and determining the multi-source session data of multiple house knowledge point labels associated therewith, the user's browsing behavior and points of interest can be captured more comprehensively. Furthermore, by generating derived multi-source session data, the feature expression data is effectively enriched, and the accuracy and depth of the analysis are improved. By extracting the session coding vector from the derived multi-source session data and combining it with the browsing path graph vector of the user's house browsing behavior big data, the accurate prediction of the user's house preference is achieved, and the corresponding house preference label data is generated, which not only improves the accuracy of house recommendations, but also provides users with a more personalized and demand-oriented house selection experience, thereby optimizing the user's browsing and selection process, and enhancing user satisfaction and platform usage efficiency.

[0103] In a possible implementation, the multi-source session data of the multiple house source knowledge point tags include at least two of the multi-source session data of house source feature identification tags, the multi-source session data of user behavior description tags, or the multi-source session data of market trend analysis tags. The multi-source session data of the house source feature identification tags is the session data of the house source features in the big data of the user's house source browsing behavior, the multi-source session data of the user behavior description tags is the session data describing the user's browsing behavior pattern in the big data of the user's house source browsing behavior, and the multi-source session data of the market trend analysis tags is the session data of the target market trend knowledge point in the big data of the user's house source browsing behavior.

[0104] In this embodiment, it is assumed that the server has received big data of house browsing behaviors of multiple users. The following description will be made using users Mr. Li, Ms. Wang, and Mr. Zhao as examples.

[0105] Mr. Li has frequently browsed houses in the past two months, and the server has obtained the big data of his house browsing behavior. The server first analyzed and found that a considerable number of the houses browsed by Mr. Li were villas located in the emerging development areas of the city. These villas have house characteristics such as intelligent home systems, private gardens and independent garages. The server identified "features of villas in emerging development areas" as a house feature identification label and obtained relevant multi-source session data, such as other users' sharing of their experience of using intelligent home systems for such villas, and experts' analysis of the potential for real estate appreciation in emerging development areas. At the same time, the server analyzed Mr. Li's browsing behavior pattern and found that when he browsed houses, he always checked the overall layout of the house first, and then focused on the surrounding commercial supporting facilities. Therefore, "user behavior description" was determined as a label, and the relevant multi-source session data included discussions by other users with similar behavior patterns on the importance of house layout and commercial supporting facilities. In addition, the server also found that Mr. Li would often check the local government's planning documents for emerging development areas during his browsing process, and pay attention to the impact of policies on housing prices in the area. Therefore, "market trend analysis" is also identified as a tag, and the server obtains relevant multi-source session data, such as the government's latest urban development planning report and professional institutions' forecasts on housing price trends in the region.

[0106] Ms. Wang also has a large number of house browsing records in the past month. The server analysis found that Ms. Wang is mainly interested in small-sized apartments in the city center, which are characterized by fine decoration and convenient transportation. The server identifies "features of small-sized apartments in the city center" as a house feature identification tag and obtains relevant multi-source session data, such as other tenants' evaluation of the quality of fine decoration of such apartments, analysis of the impact of transportation stations on living convenience, etc. Through the analysis of Ms. Wang's browsing behavior, the server found that every time she browses, she first checks the property management services of the community and then pays attention to the surrounding educational resources. The "user behavior description" tag is determined, and the relevant multi-source session data covers other users' exchanges on the importance of community property management and educational resources. In addition, the server noticed that Ms. Wang often reads news reports on fluctuations in housing prices in the city center and pays attention to changes in market supply and demand. The "market trend analysis" tag is enabled, and the server obtains relevant multi-source session data, including recent transaction data of the real estate market in the city center, policy control measures on housing prices in the city center, etc.

[0107] Mr. Zhao's browsing behavior in the past three weeks has attracted the attention of the server. The server determined that Mr. Zhao mainly browsed large-sized houses in the suburbs, which have spacious balconies, underground storage rooms and other house characteristics. "Features of large-sized houses in the suburbs" became the house feature identification label, and the multi-source session data obtained by the server included the owner's discussion on how to use the spacious balcony and the introduction of moisture-proof measures for the underground storage room. The server further analyzed Mr. Zhao's browsing behavior and found that he always checked the lighting of the house first, and then paid attention to the leisure facilities in the community. The "User Behavior Description" label was established, and the relevant multi-source session data included other home buyers' views on the importance of house lighting and community leisure facilities. At the same time, the server found that Mr. Zhao would compare the price trends of different suburban properties and was concerned about the role of regional development in improving the value of real estate. The "Market Trend Analysis" label was applied, and the server obtained multi-source session data such as new construction project planning in the suburbs and historical housing price change data.

[0108] Through the above analysis of the big data of Mr. Li, Ms. Wang and Mr. Zhao's house browsing behavior, the server successfully determined the multi-source session data including house feature identification tags, user behavior description tags and market trend analysis tags, providing a rich and targeted data foundation for subsequent processing and analysis.

[0109] In a possible implementation, step S110 includes:

[0110] Step A110, identifying the user's house browsing behavior big data, generating target market trend knowledge points in the user's house browsing behavior big data and association information of the target market trend knowledge points in the user's house browsing behavior big data.

[0111] Step A120, obtaining reference knowledge data associated with the target market trend knowledge point from a set knowledge base, wherein the set knowledge base includes reference knowledge data for a plurality of market trend knowledge points.

[0112] Step A130, obtaining restriction information corresponding to the user's house browsing behavior big data, wherein the restriction information is used to limit the data volume of the session data describing the user's browsing behavior pattern in the user's house browsing behavior big data.

[0113] Step A140, generating session data describing the user browsing behavior pattern in the user housing browsing behavior big data based on the restriction information according to the reference knowledge data associated with the target market trend knowledge point and the associated information.

[0114] In detail, the server received the big data of the user Mr. Chen's house browsing behavior. Mr. Chen frequently browsed house information in different cities in the past quarter.

[0115] The server first identifies Mr. Chen's user house browsing behavior big data. By analyzing factors such as the types of houses that Mr. Chen browsed, the price range, and the areas he was interested in, it was found that Mr. Chen was particularly concerned about the housing price trends in emerging areas of first-tier cities. Therefore, the target market trend knowledge point "House price trends in emerging areas of first-tier cities" was generated, as well as its related information in Mr. Chen's browsing behavior big data, such as the frequency and duration of Mr. Chen's browsing of such houses.

[0116] Next, the server obtains reference knowledge data associated with the target market trend knowledge point of "housing price trends in emerging areas of first-tier cities" from the set knowledge base. The set knowledge base contains a large amount of reference knowledge about various market trend knowledge points, such as detailed reports and analysis on land supply in emerging areas of first-tier cities, infrastructure construction planning, and the impact of policies on the real estate market.

[0117] Then, the server obtains the restriction information corresponding to Mr. Chen's user house browsing behavior big data. These restriction information may be caused by factors such as system settings, data storage capacity or data processing capacity, and are used to limit the amount of data describing the session data of Mr. Chen's user browsing behavior pattern. Assume that the restriction information stipulates that the session data describing the user browsing behavior pattern cannot exceed 1000 bytes.

[0118] Finally, the server generates descriptive session data of Mr. Chen's user browsing behavior pattern based on the obtained reference knowledge data associated with the target market trend knowledge point and the previously generated associated information, based on the restriction information. For example, the generated descriptive session data may be: "In the past quarter, Mr. Chen has frequently paid attention to the listings in the emerging areas of first-tier cities, and has shown a strong interest in the housing price trend. He browses such listings an average of 5 times a week, and stays for about 10 minutes each time. Referring to the information in the knowledge base, the land supply in this area has increased recently, infrastructure construction has accelerated, and policy regulation tends to support housing demand. It is expected that housing prices will maintain a stable upward trend."

[0119] Assume that the server receives the big data of the house browsing behavior of user Ms. Lin. In the past two months, Ms. Lin mainly browsed the relevant information of school district houses in second-tier cities.

[0120] The server identifies Ms. Lin's browsing behavior big data, determines "price fluctuations of school district housing in second-tier cities" as the target market trend knowledge point, and obtains its related information in the browsing behavior big data, such as the frequency of Ms. Lin's viewing of prices of different school district housing, detailed records of price comparisons, etc.

[0121] From the set knowledge base, the server obtains reference knowledge data related to "fluctuations in school district housing prices in second-tier cities", including changes in the distribution of local educational resources, the impact of adjustments to school admission policies on school district housing prices, etc.

[0122] Restriction information is obtained, such as the description of Ms. Lin’s user browsing behavior pattern, and the session data is limited to 800 bytes.

[0123] Based on the above reference knowledge data, related information and restriction information, the description session data is generated: "Ms. Lin has focused on browsing school district housing in second-tier cities in the past two months, checking prices about 3 times a week, and each time for an average of 8 minutes. The knowledge base shows that local high-quality educational resources have expanded to new districts, and some old school districts have adjusted their enrollment policies, resulting in a decrease in the prices of some school district housing, while the prices of school district housing in new districts are on an upward trend."

[0124] Assume that the server receives the big data of the house browsing behavior of user Mr. Wu. In the past month, Mr. Wu has focused on browsing the information of sea view houses in tourist cities.

[0125] The server identifies "Market popularity of sea view housing in tourist cities" as a target market trend knowledge point and related associated information, such as Mr. Wu's attention level and time distribution to different sea view housing projects.

[0126] Obtain reference knowledge related to target market trend knowledge points such as tourism development plans of tourist cities and sea view housing development policies from the knowledge base.

[0127] Obtain restricted information, such as the description of Mr. Wu’s user browsing behavior pattern. The session data cannot exceed 600 bytes.

[0128] The generated description session data is: "Mr. Wu has frequently paid attention to sea view houses in tourist cities in the past month. He browses them once every three days, and each time for about 5 minutes. The knowledge base shows that the tourist city has increased tourism promotion, the development of sea view houses is limited, the market popularity has increased, but the supply is limited, and the price has risen steadily."

[0129] In a possible implementation, step S110 includes:

[0130] Step B110, performing behavior pattern recognition on the user's house browsing behavior big data, and generating concerned house data, core browsing data and auxiliary browsing data in the user's house browsing behavior big data.

[0131] Step B120: extracting a property vector from the property data, extracting a core browsing vector from the core browsing data, and extracting an auxiliary browsing vector from the auxiliary browsing data.

[0132] Step B130, generating session data describing the user browsing behavior pattern in the user's house browsing behavior big data based on the concerned house vector, the core browsing vector and the auxiliary browsing vector.

[0133] In this embodiment, the server receives the big data of the house browsing behavior of the user Mr. Liu, who has browsed houses many times in the past two months.

[0134] The server first identified the behavior patterns of Mr. Liu's user listing browsing behavior big data. By analyzing factors such as the duration of each browsing, the frequency of clicks, and the number of repeated views, the following data was determined:

[0135] Mr. Liu paid special attention to three-bedroom houses located in the city center, close to key schools and with gardens, and identified these as the houses of interest. For example, he checked the detailed information of a 120-square-meter three-bedroom house in a community many times, including the floor plan, introduction of surrounding schools, etc. This is a typical house of interest.

[0136] At the same time, the server identified the information Mr. Liu focused on during his browsing, such as house prices, house orientation, and community supporting facilities, which constituted the core browsing data. For example, he carefully studied the comparison chart of house prices, had a clear preference for houses with north-south orientation, and checked many times whether there were supporting facilities such as gyms and supermarkets in the community.

[0137] Some less critical but relevant information, such as the construction year of the house, the developer's brand, etc., are classified as auxiliary browsing data.

[0138] Next, the server extracts the vector of the house of interest from the house of interest data. For the three-bedroom house mentioned above, a vector such as [1, 0, 0] may be extracted, where 1 represents a three-bedroom house and 0 represents other house types.

[0139] Extract core browsing vectors from core browsing data. For example, for the preference for housing prices between 2 million and 3 million, extract the vectors [0.8, 0.2, 0], which represent the preference for low, medium, and high price ranges, respectively. For the preference for north-south transparent orientation, extract the vectors [1, 0, 0], which represent the preference for different orientations.

[0140] Extract auxiliary browsing vectors from auxiliary browsing data. For example, for the acceptance level of houses built between 5 and 10 years old, extract the vector [0.6, 0.4, 0].

[0141] Finally, the server generates session data describing Mr. Liu's browsing behavior pattern based on the vector of the listing he is interested in, the core browsing vector, and the auxiliary browsing vector. For example, the following description is generated: "Mr. Liu mainly focuses on three-bedroom apartments in the city center when browsing listings. He has a clear preference for houses with a price range of 2-3 million yuan and a north-south orientation. At the same time, he is also somewhat receptive to houses built 5-10 years ago. He focuses on checking the supporting facilities in the community, especially gyms and supermarkets."

[0142] Assume that the server receives the big data of the house browsing behavior of user Ms. Zhou.

[0143] The server analyzes Ms. Zhou's browsing behavior and determines that she is interested in villas in the suburbs, which is considered as the data of interested properties. For example, she has viewed the detailed description of a villa with a private swimming pool in a suburb many times.

[0144] The core browsing data includes the villa's area, decoration style, surrounding natural environment, etc. She carefully compared villas of different sizes, showed interest in European-style decoration style, and paid close attention to whether there were natural landscapes such as parks and lakes around.

[0145] Auxiliary browsing data such as the villa's property management fees, number of parking spaces, etc.

[0146] The server extracts the vector [0, 0, 1] from the concerned house data to represent villas. The vector [0.7, 1, 0.9] is extracted from the core browsing data to represent the preference for area, decoration style, and surrounding natural environment. The vector [0.5, 0.5] is extracted from the auxiliary browsing data to represent the attention paid to property management fees and the number of parking spaces.

[0147] The resulting description session data is: "Ms. Zhou focuses on villas in the suburbs when browsing properties. She prefers villas with larger areas, European-style decoration, and good natural environment. She is also concerned about property management fees and the number of parking spaces."

[0148] In a possible implementation, step S130 further includes:

[0149] Step S131, integrating the multi-source conversation knowledge vector sequence of the multiple house resource knowledge point labels to generate a multi-source conversation knowledge vector integrated sequence.

[0150] Step S132: Process the multi-source conversation knowledge vector integration sequence using a self-attention mechanism to determine a self-attention coefficient of the multi-source conversation knowledge vector integration sequence.

[0151] Step S133: update the knowledge vector value of the multi-source conversation knowledge vector integration sequence according to the self-attention coefficient of the multi-source conversation knowledge vector integration sequence, and generate derivative multi-source conversation data of the user's house browsing behavior big data according to the updated knowledge vector value.

[0152] In this embodiment, the server obtains a multi-source conversation knowledge vector sequence of multiple housing knowledge point tags of the user Mr. Zhang, including vector sequences corresponding to the three tags of "city center housing features", "market price trends" and "user own needs".

[0153] First, the server integrates the multi-source conversational knowledge vector sequences of the three property knowledge point labels. Assume that the vector sequence of "city center property features" is [0.8, 0.6, 0.4], the vector sequence of "market price trends" is [0.7, 0.5, 0.3], and the vector sequence of "user needs" is [0.9, 0.7, 0.2]. The server integrates them to generate a multi-source conversational knowledge vector integration sequence, such as [0.8, 0.6, 0.4, 0.7, 0.5, 0.3, 0.9, 0.7, 0.2].

[0154] Next, the server uses the self-attention mechanism to process this integrated sequence of multi-source conversation knowledge vectors. Through a series of complex calculations, the self-attention coefficient of the sequence is determined. For example, it is calculated that in this integrated sequence, the self-attention coefficient of the part related to "city center property features" is 0.4, the self-attention coefficient of the part related to "market price trends" is 0.3, and the self-attention coefficient of the part related to "user's own needs" is 0.3.

[0155] Then, the server updates the knowledge vector values ​​of the multi-source conversation knowledge vector integration sequence based on these self-attention coefficients. For example, the vector value corresponding to "city center property features" is multiplied by 0.4, the vector value corresponding to "market price trends" is multiplied by 0.3, and the vector value corresponding to "user's own needs" is multiplied by 0.3.

[0156] Finally, the derived multi-source session data of Mr. Zhang’s user house browsing behavior big data is generated based on the updated knowledge vector value. For example, the following session data is generated: “Mr. Zhang is more concerned about the characteristics of downtown houses, and also considers the market price trend and his own needs, and tends to look for downtown houses with stable prices and that meet his own needs.”

[0157] Assume again that the server obtains a multi-source conversation knowledge vector sequence of multiple housing knowledge point labels of user Ms. Wang, namely "characteristics of school district housing", "regional development prospects" and "family living needs".

[0158] The server integrates these three vector sequences. Assuming that "characteristics of school district housing" is [0.7, 0.4, 0.3], "regional development prospects" is [0.6, 0.5, 0.2], and "family living needs" is [0.8, 0.6, 0.1], the integrated sequence is [0.7, 0.4, 0.3, 0.6, 0.5, 0.2, 0.8, 0.6, 0.1].

[0159] After processing using the self-attention mechanism, the self-attention coefficient of "characteristics of school district housing" was determined to be 0.3, the self-attention coefficient of "regional development prospects" was 0.4, and the self-attention coefficient of "family living needs" was 0.3.

[0160] After updating the knowledge vector value based on the self-attention coefficient, the derived multi-source conversation data is generated: "Ms. Wang focuses on the characteristics of school district housing when choosing a house. She also fully considers the development prospects of the region and the family's living needs, hoping to find a suitable house that can meet the needs of her children's education and family life."

[0161] In a possible implementation, the browsing path graph vector includes the browsing path node vector of the user's house browsing behavior big data. Step S140 includes:

[0162] Step S141, decomposing the user's house browsing behavior big data into multiple house browsing behavior paths.

[0163] Step S142, constructing a behavior pattern association graph based on the multiple property browsing behavior paths, the behavior pattern association graph includes multiple graph nodes and multiple association links, the multiple graph nodes are used to reflect different property browsing behavior paths, and the multiple association links are used to reflect the behavior pattern association information between the property browsing behavior paths.

[0164] Step S143, performing graph convolution processing on the behavior pattern association graph to generate a browsing path node vector corresponding to the behavior pattern association graph.

[0165] In this embodiment, the server receives the big data of the house browsing behavior of the user Mr. Li, who has browsed houses many times in the past month.

[0166] The server first decomposes Mr. Li's user house browsing behavior big data into multiple house browsing behavior paths. For example, Mr. Li first browsed a three-bedroom apartment in Community A in the city center, then checked a villa in Community B in the suburbs, and then returned to the city center to check a two-bedroom apartment in Community C. In this way, three house browsing behavior paths were decomposed: three-bedroom apartment in Community A, villa in Community B, and two-bedroom apartment in Community C.

[0167] Next, the server constructs a behavior pattern association graph based on these multiple house browsing behavior paths. In this graph, the three-bedroom apartment in Community A, the villa in Community B, and the two-bedroom apartment in Community C are each a graph node. In terms of association links, if Mr. Li views the villa in Community B immediately after viewing the three-bedroom apartment in Community A, then there is an association link from the three-bedroom apartment in Community A to the villa in Community B, reflecting Mr. Li's browsing behavior pattern association information from downtown housing to suburban housing. Similarly, if Mr. Li returns to view the two-bedroom apartment in Community C after viewing the villa in Community B, there will be a corresponding association link.

[0168] The server then performs graph convolution on the behavior pattern association graph. Through complex mathematical operations, the association relationship and weight between each node are analyzed to generate a browsing path node vector corresponding to the behavior pattern association graph. For example, the generated vector may reflect that Mr. Li pays more attention to downtown properties, has relatively less interest in suburban properties, and has a certain tendency to browse and switch between different types of properties.

[0169] Assume that the server receives the big data of the user Ms. Zhao's house browsing behavior. Ms. Zhao's browsing behavior in the past two weeks is as follows: first, she viewed a small apartment in the D community near the school in the city, then browsed a large apartment in the E community near the commercial center in the city, and then paid attention to a vacation home in the F community in the suburbs with beautiful scenery.

[0170] The server breaks down Ms. Zhao's browsing behavior into three housing browsing behavior paths: small apartments in Community D, large apartments in Community E, and vacation homes in Community F.

[0171] When constructing the behavior pattern association graph, these three house browsing behavior paths are used as graph nodes. If Ms. Zhao first looks at the small apartment in D Community and then the large apartment in E Community, there is an association link from the small apartment node in D Community to the large apartment node in E Community, indicating that Ms. Zhao has shifted her focus from the small apartment near the school district to the large apartment near the commercial center. If she then looks at the vacation home in F Community, there will be another corresponding link.

[0172] After performing graph convolution processing on this behavior pattern association graph, the generated browsing path node vector may indicate that Ms. Zhao prefers houses in the urban area, has relatively little interest in vacation homes, and has a specific pattern in browsing decisions between houses with different functions and locations.

[0173] In a possible implementation, step S142 includes:

[0174] Step S1421, respectively identifying target behavior pattern features in each of the house browsing behavior paths.

[0175] Step S1422, constructing a behavior pattern association graph based on the target behavior pattern features in the multiple property browsing behavior paths, wherein the multiple graph nodes of the behavior pattern association graph are used to reflect the target behavior pattern features in different property browsing behavior paths, and the multiple association links are used to reflect the behavior pattern association information between the target behavior pattern features.

[0176] In this embodiment, the server obtains multiple house browsing behavior paths of user Mr. Sun. Mr. Sun's browsing behavior in the past three weeks includes: first browsing a high-end apartment in G Community, located in the core business district of the city, which features convenient transportation and complete supporting facilities; then checking a flat house in H Community, located in the emerging science and technology park of the city, which has the advantage of great future development potential; then paying attention to a single-family villa in I Community, located in the suburbs of the city, which has a beautiful environment and spacious space.

[0177] The server identifies the target behavior pattern features in each property browsing behavior path. For the high-end apartments in Community G, the target behavior pattern features are "convenient transportation and complete supporting facilities"; for the flat houses in Community H, the feature is "great development potential"; for the single-family villas in Community I, the feature is "good environment and large space".

[0178] Then, a behavior pattern association graph is constructed based on these target behavior pattern characteristics. In this graph, "convenient transportation and complete supporting facilities", "great development potential", and "good environment and large space" are used as three graph nodes. In terms of association links, if Mr. Sun browses the house with the feature of "convenient transportation and complete supporting facilities" and then browses the house with the feature of "great development potential", then there is an association link from the "convenient transportation and complete supporting facilities" node to the "great development potential" node, reflecting Mr. Sun's behavior pattern association information from focusing on current convenient conditions to future development potential. Similarly, if Mr. Sun pays attention to the house with "good environment and large space" after viewing the house with "great development potential", there will be a corresponding association link.

[0179] Assume that the server has obtained the browsing behavior path of user Ms. Zhou. Ms. Zhou's browsing behavior in the past month is as follows: first, she looked at an old school district house in J community near a famous school in the city, focusing on high-quality educational resources; then she looked at a lake view house in K community located by the city lake, with the outstanding feature of charming scenery; then she paid attention to a renovated small apartment in L community in the old city, with the advantages of excellent geographical location and affordable price.

[0180] The server identifies the target behavior pattern features in the browsing behavior paths of these properties. The feature of the old school district house in J community is "high-quality educational resources", the feature of the lake view house in K community is "charming scenery", and the feature of the renovated small apartment in L community is "good location and low price".

[0181] A behavior pattern association graph is constructed based on these target behavior pattern characteristics. In the graph, "high-quality educational resources", "charming landscape", and "good location and low price" are used as graph nodes. If Ms. Zhou browses the listings of "high-quality educational resources" and then "charming landscape", there will be a corresponding association link, reflecting her change in behavior pattern from focusing on educational resources to pursuing landscape enjoyment. When she pays attention to the listings of "good location and low price", a corresponding link will also be formed.

[0182] In a possible implementation, the browsing path graph vector further includes a housing location association vector of the user housing browsing behavior big data, and the method further includes:

[0183] Step C110, constructing a location association graph based on the multiple property browsing behavior paths, the location association graph includes multiple graph nodes and multiple association links, the multiple graph nodes are used to reflect different property browsing behavior paths, and the multiple association links are used to reflect the property location association information between the property browsing behavior paths.

[0184] Step C120, performing graph convolution processing on the location association graph to generate a house location association vector corresponding to the location association graph.

[0185] In this embodiment, the server obtains multiple house browsing behavior paths of user Mr. Wu. Mr. Wu's browsing behavior in the past two months is as follows: first, he browsed a high-rise residential building in M ​​community in the new city in the east of the city, then he viewed a bungalow in N community in the old city in the west of the city, and then he paid attention to an apartment in O community near the university town in the south of the city.

[0186] The server constructs a location association graph based on these multiple house browsing behavior paths. In this graph, "M Community in the new city in the east", "N Community in the old city in the west", and "O Community near the university town in the south of the city" are respectively used as graph nodes. In terms of association links, if Mr. Wu browses the M Community in the new city in the east and then checks the N Community in the old city in the west, there will be an association link from the "M Community in the new city in the east" node to the "N Community in the old city in the west", reflecting Mr. Wu's location association information from the east to the west of the city. Similarly, if Mr. Wu pays attention to the O Community near the southern university town after checking the N Community in the old city in the west, there will be a corresponding association link.

[0187] The server then performs graph convolution on the location association graph. Through a series of mathematical operations and analyses, it generates a property location association vector corresponding to the location association graph. For example, the generated vector may indicate Mr. Wu’s attention and preference for different areas of the city, and may show that he is more concerned about the location of properties near the new city and university town.

[0188] Assume that the server obtains the browsing behavior paths of multiple listings of the user Ms. Zheng. Ms. Zheng's browsing behavior in the past month and a half is as follows: first, she viewed a commercial house in P Community near the industrial park in the north of the city, then browsed an office building converted into a residential building in Q Community in the central business district of the city, and then paid attention to a villa in R Community next to the forest park in the suburbs of the city.

[0189] The server builds a location association graph based on these house browsing behavior paths. In the graph, "Community P near the industrial park in the north of the city", "Community Q in the central business district of the city", and "Community R next to the forest park in the suburbs of the city" are graph nodes. If Ms. Zheng first browses Community P near the industrial park in the north, and then visits Community Q in the central business district, there will be a corresponding association link, indicating her location change from the north of the city to the center. When she pays attention to Community R next to the forest park in the suburbs, a new link will be formed.

[0190] Afterwards, the server performs graph convolution processing on this location association graph, and the generated property location association vector may reflect Ms. Zheng’s preference for specific locations in the city center and suburbs, as well as the association pattern of her browsing behavior between different locations.

[0191] In a possible implementation, step S150 includes:

[0192] Step S151, fusing the derived session coding vector and the browsing path graph vector to generate fused vector data.

[0193] Step S152: input the fused vector data into a pre-trained housing preference prediction model to perform housing preference prediction, and generate housing preference label data corresponding to the user housing browsing behavior big data.

[0194] In this embodiment, it is assumed that the server obtains the derived session coding vector and browsing path map vector of user Mr. Chen. Mr. Chen's derived session coding vector reflects his attention and preference for house prices, house types, decoration styles, etc., for example, the coding vector is [0.7, 0.2, 0.1, 0.05]. His browsing path map vector reflects his browsing order and attention between different areas and different types of houses, for example, [0.6, 0.3, 0.1].

[0195] The server first merges the two vectors to generate merged vector data. For example, by using a specific algorithm, the two vectors are added together to obtain merged vector data [1.3, 0.5, 0.2, 0.15].

[0196] The server then inputs the fused vector data into a pre-trained property preference prediction model for property preference prediction. This property preference prediction model has been trained with a large amount of user data and is able to understand and analyze the meaning of the fused vector data.

[0197] After calculation and analysis by the model, the house preference label data corresponding to Mr. Chen's user house browsing behavior big data was generated. The prediction result may show that Mr. Chen prefers houses located in the city center, moderately priced, medium-sized, and modern and simple decoration. The corresponding house preference label data may be "city center, moderately priced, medium-sized, modern and simple decoration".

[0198] Assume that the server processes the case of user Ms. Lin. Ms. Lin's derived session encoding vector is [0.8, 0.1, 0.08, 0.02], and the browsing path graph vector is [0.7, 0.2, 0.1].

[0199] The server merges the two and gets [1.5, 0.3, 0.18, 0.12].

[0200] The fused vector data is input into the pre-trained housing preference prediction model. The housing preference prediction model predicts that Ms. Lin prefers houses located in the suburbs, with low prices, large units and gardens. The generated housing preference label data may be "suburbs, low prices, large units and gardens".

[0201] Further, Figure 2 FIG. 1 is a schematic diagram showing a hardware structure of a housing recommendation system 100 for implementing the method provided in an embodiment of the present application. Figure 2 As shown, the house recommendation system 100 may include at least one processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 and a controller 108 for communication functions. It can be understood by those skilled in the art that Figure 2 The structure shown is for illustration only and does not limit the structure of the house recommendation system 100. For example, the house recommendation system 100 may also include Figure 2 More or fewer components as shown, or with Figure 2 Different configurations are shown.

[0202] The memory 104 can be used to store software programs and modules of application software, such as program instructions corresponding to the above-mentioned method embodiments in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned user behavior big data analysis method based on artificial intelligence. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the house recommendation system 100 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0203] The transmission device 106 is used to obtain or send data via a network. A specific example of the above network may include a wireless network provided by a communication provider of the house recommendation system 100. In one example, the transmission device 106 includes a network adapter, which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency module, which is used to communicate with the Internet wirelessly.

[0204] It should be noted that the order of the above embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of the embodiments of the present application. Other embodiments are within the scope of the attached claims. In some cases, the exceptions or steps recorded in the claims can be performed according to an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or can be advantageous.

[0205] Each embodiment in the embodiments of the present application is described in a progressive manner, and the parts that are consistent and similar between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above different embodiments, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0206] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program. The above program may be stored in a computer-readable storage medium, and the above storage medium may be a read-only memory, a disk or an optical disk, etc.

Claims

1. A user behavior big data analysis method based on artificial intelligence, characterized in that: The method comprises: Obtaining user house browsing behavior big data, and determining multi-source session data of multiple house knowledge point tags associated with the user house browsing behavior big data; Determine a multi-source conversation knowledge vector sequence of the multiple housing source knowledge point labels according to the multi-source conversation data of the multiple housing source knowledge point labels; Based on the multi-source conversation knowledge vector sequences of the multiple house knowledge point labels, generating derived multi-source conversation data of the user's house browsing behavior big data, wherein the derived multi-source conversation data includes more feature expression data than the multi-source conversation data of the multiple house knowledge point labels; Extracting a session coding vector from the derived multi-source session data to generate a derived session coding vector; Extracting a browsing path graph vector from the user's house browsing behavior big data; According to the derived session coding vector and the browsing path graph vector, predicting the housing preference of the user's housing browsing behavior big data, and generating housing preference label data corresponding to the user's housing browsing behavior big data; The multi-source session data of the plurality of house source knowledge point labels include at least two of the multi-source session data of house source feature identification labels, the multi-source session data of user behavior description labels, or the multi-source session data of market trend analysis labels; wherein the multi-source session data of the house source feature identification labels is the session data of the house source features in the big data of the user's house source browsing behavior, the multi-source session data of the user behavior description labels is the session data describing the user's browsing behavior pattern in the big data of the user's house source browsing behavior, and the multi-source session data of the market trend analysis labels is the session data of the target market trend knowledge point in the big data of the user's house source browsing behavior; The multi-source session data of determining a plurality of property knowledge point tags associated with the user property browsing behavior big data includes: Performing behavior pattern recognition on the user's house browsing behavior big data to generate concerned house data, core browsing data and auxiliary browsing data in the user's house browsing behavior big data; Extracting a focused property vector from the focused property data, extracting a core browsing vector from the core browsing data, and extracting an auxiliary browsing vector from the auxiliary browsing data; Generate session data describing the user browsing behavior pattern in the user's house browsing behavior big data based on the concerned house vector, the core browsing vector and the auxiliary browsing vector; The generating of the derived multi-source session data of the user's house browsing behavior big data based on the multi-source session knowledge vector sequence of the multiple house knowledge point tags also includes: Integrate the multi-source conversation knowledge vector sequence of the multiple house knowledge point labels to generate a multi-source conversation knowledge vector integration sequence; Processing the multi-source conversation knowledge vector integration sequence using a self-attention mechanism to determine a self-attention coefficient of the multi-source conversation knowledge vector integration sequence; updating the knowledge vector value of the multi-source conversation knowledge vector integration sequence according to the self-attention coefficient of the multi-source conversation knowledge vector integration sequence, and generating derivative multi-source conversation data of the user's house browsing behavior big data according to the updated knowledge vector value; The browsing path graph vector includes the browsing path node vector of the user's house browsing behavior big data; the step of extracting the browsing path graph vector from the user's house browsing behavior big data includes: Decomposing the user's house browsing behavior big data into multiple house browsing behavior paths; Constructing a behavior pattern association graph according to the plurality of house browsing behavior paths, the behavior pattern association graph comprising a plurality of graph nodes and a plurality of association links, the plurality of graph nodes being used to reflect different house browsing behavior paths, and the plurality of association links being used to reflect behavior pattern association information between the respective house browsing behavior paths; Performing graph convolution processing on the behavior pattern association graph to generate a browsing path node vector corresponding to the behavior pattern association graph; The constructing a behavior pattern association graph based on the multiple house browsing behavior paths includes: Respectively identifying target behavior pattern features in each of the house browsing behavior paths; Constructing a behavior pattern association graph according to the target behavior pattern features in the multiple house browsing behavior paths, wherein the multiple graph nodes of the behavior pattern association graph are used to reflect the target behavior pattern features in different house browsing behavior paths, and the multiple association links are used to reflect the behavior pattern association information between the target behavior pattern features; The browsing path graph vector also includes a housing location association vector of the user's housing browsing behavior big data, and the method further includes: Constructing a location association graph according to the plurality of house browsing behavior paths, the location association graph comprising a plurality of graph nodes and a plurality of association links, the plurality of graph nodes being used to reflect different house browsing behavior paths, and the plurality of association links being used to reflect the house location association information between the house browsing behavior paths; Perform graph convolution processing on the location association graph to generate a house location association vector corresponding to the location association graph.

2. The method for analyzing user behavior big data based on artificial intelligence according to claim 1 is characterized in that: The method of predicting housing preference of the user's housing browsing behavior big data based on the derived session coding vector and the browsing path graph vector, and generating housing preference label data corresponding to the user's housing browsing behavior big data, includes: Merging the derived session encoding vector and the browsing path graph vector to generate fused vector data; The fused vector data is input into a pre-trained housing preference prediction model to perform housing preference prediction, and generate housing preference label data corresponding to the user housing browsing behavior big data.

3. A house recommendation system, characterized in that: The housing recommendation system includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the user behavior big data analysis method based on artificial intelligence according to any one of claims 1-2 is implemented.

Images