House recommendation processing method and system based on big data

Abstract

The embodiment of the present application provides a method and system for processing house recommendation based on big data. The embodiment of the present application realizes efficient and accurate house recommendation by finely processing the big data of house behavior of target users. Specifically, the target house behavior big data is first decomposed to generate multiple house behavior data blocks, and then the associated path information of these blocks is obtained and encoded, and a network knowledge graph containing rich network member information is constructed. Based on this network knowledge graph, the target house behavior big data can be deeply mined to generate a recommended keyword sequence that reflects user needs, thereby realizing highly personalized house recommendation for target users, significantly improving the accuracy of house recommendation and user satisfaction.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to a method and system for processing house recommendation based on big data. Background Art

[0002] With the rapid development of Internet technology, the real estate industry has gradually shifted from the traditional offline model to the online model, and the way of obtaining and recommending real estate information has also undergone profound changes. Traditional housing recommendation methods mostly rely on manual screening or simple search matching, which is difficult to accurately capture the real needs of users, resulting in unsatisfactory recommendation results. Especially in the era of big data, user behavior data is massive, diverse, and growing rapidly. How to effectively use these big data to recommend housing has become a problem that the real estate industry needs to solve.

[0003] Although the existing house recommendation system has begun to try to introduce big data technology, there are still many deficiencies in data processing and recommendation algorithms. On the one hand, traditional data processing methods often ignore the complex correlation and dynamic changes between user behavior data, and it is difficult to fully and accurately reflect the real needs of users; on the other hand, the existing recommendation algorithms are mostly based on simple similarity calculations or rule matching, lacking deep semantic understanding and context perception capabilities, resulting in inaccurate and in personalized recommendation results. Summary of the invention

[0004] In view of this, the purpose of this application is to provide a housing recommendation processing method and system based on big data.

[0005] According to a first aspect of the present application, a method for processing house recommendation based on big data is provided, which is applied to a house recommendation system, and the method comprises:

[0006] Obtaining target housing behavior big data of a target user, and decomposing the target housing behavior big data according to the interactive behaviors included in the target housing behavior big data to generate X housing behavior data blocks of the target housing behavior big data;

[0007] Obtaining associated path information of each of the X housing source behavior data blocks in the target housing source behavior big data, encoding and representing the X housing source behavior data blocks, and generating encoding and representing vector data of each of the X housing source behavior data blocks;

[0008] For each of the X housing source behavior data blocks, the housing source behavior data block is used as a network member of the network knowledge graph, and network member information of the network member is determined according to the encoding representation vector data and the associated path information of the housing source behavior data block, so as to generate the network knowledge graph of the target housing source behavior big data;

[0009] According to the network knowledge graph of the target housing source behavior big data, data mining is performed on the target housing source behavior big data to generate a recommended keyword sequence, and housing sources are recommended to the target user based on the recommended keyword sequence.

[0010] In a possible implementation manner of the first aspect, determining the network member information of the network member according to the coded representation vector data and the associated path information of the housing source behavior data block includes:

[0011] Obtaining behavior pattern statistics of the housing source behavior data block;

[0012] The network member information of the network members is determined based on the coded representation vector data, associated path information and behavior pattern statistical data of the housing source behavior data block.

[0013] In a possible implementation of the first aspect, performing data mining on the target property behavior big data to generate a recommended keyword sequence based on a network knowledge graph of the target property behavior big data includes:

[0014] Determine a recommended knowledge base, the recommended knowledge base comprising a recommended keyword weight distribution of each first template property behavior data in Y first template property behavior data, and a knowledge graph vector sequence or a network knowledge graph of each first template property behavior data, wherein the knowledge graph vector sequence of the first template property behavior data is determined based on the network knowledge graph of the first template property behavior data, and Y is a positive integer greater than 1;

[0015] Determine a matching score between the target property behavior big data and each of the first template property behavior data according to the network knowledge graph of the target property behavior big data and the knowledge graph vector sequence or network knowledge graph of each of the first template property behavior data;

[0016] Determine Z first template property listing behavior data from the Y first template property listing behavior data according to the matching score, where Z is not greater than Y;

[0017] Generate a recommended keyword sequence according to the recommended keyword weight distribution of the Z first template housing source behavior data.

[0018] In a possible implementation of the first aspect, determining the recommended knowledge base includes:

[0019] Obtaining the Y first template house listing behavior data;

[0020] Determine a network knowledge graph and a recommended keyword weight distribution of each first template housing source behavior data in the Y first template housing source behavior data;

[0021] The recommended knowledge base is determined based on the network knowledge graph of the Y first template housing source behavior data and the weight distribution of recommended keywords.

[0022] For example, in a possible implementation of the first aspect, obtaining the Y first template housing source behavior data includes:

[0023] Obtain S reference template property listing behavior data and T candidate template property listing behavior data, where S and T are both positive integers;

[0024] Based on the S reference template housing source behavior data, at least one is selected from the T candidate template housing source behavior data as the first template housing source behavior data.

[0025] For example, in a possible implementation of the first aspect, selecting at least one of the S reference template housing source behavior data from the T candidate template housing source behavior data as the first template housing source behavior data includes:

[0026] For the i-th candidate template property listing behavior data among the T candidate template property listing behavior data, determine a matching score between the i-th candidate template property listing behavior data and each reference template property listing behavior data among the S reference template property listing behavior data, where i is a positive integer not greater than T;

[0027] If the matching score between the i-th candidate template housing source behavior data and at least one of the S reference template housing source behavior data is greater than a second threshold value, the i-th candidate template housing source behavior data is output as the first template housing source behavior data.

[0028] For example, in a possible implementation of the first aspect, determining a matching score between the i-th candidate template housing source behavior data and each reference template housing source behavior data in the S reference template housing source behavior data includes:

[0029] Determine the knowledge graph vector sequence of the i-th candidate template property behavior data and the knowledge graph vector sequence of each reference template property behavior data in the S reference template property behavior data;

[0030] Based on the knowledge graph vector sequence of the i-th candidate template property behavior data and the knowledge graph vector sequence of each reference template property behavior data in the S reference template property behavior data, determine the matching score between the i-th candidate template property behavior data and each reference template property behavior data in the S reference template property behavior data.

[0031] For example, in a possible implementation of the first aspect, determining the knowledge graph vector sequence of the i-th candidate template property behavior data and the knowledge graph vector sequence of each reference template property behavior data in the S reference template property behavior data includes:

[0032] Determine a network knowledge graph of the i-th candidate template property listing behavior data and a network knowledge graph of each reference template property listing behavior data in the S reference template property listing behavior data;

[0033] Determine a knowledge graph vector sequence of the i-th candidate template housing source behavior data according to the network knowledge graph of the i-th candidate template housing source behavior data;

[0034] For each reference template housing source behavior data in the S reference template housing source behavior data, a knowledge graph vector sequence of the reference template housing source behavior data is determined based on the network knowledge graph of the reference template housing source behavior data.

[0035] For example, in a possible implementation of the first aspect, determining the recommended keyword weight distribution of each of the Y first template housing source behavior data includes:

[0036] If the first template housing source behavior data is the i-th candidate template housing source behavior data, then according to the matching score between the i-th candidate template housing source behavior data and each reference template housing source behavior data in the S reference template housing source behavior data, T target template housing source behavior data whose matching score with the i-th candidate template housing source behavior data is not less than a third threshold value are determined from the S reference template housing source behavior data;

[0037] According to the recommended keyword weight distributions of the T candidate template housing source behavior data, the recommended keyword weight distribution of the i-th candidate template housing source behavior data is determined.

[0038] For example, in a possible implementation of the first aspect, determining a network knowledge graph of the third template housing source behavior data includes:

[0039] Decomposing the third template housing source behavior data according to the interactive behaviors included in the third template housing source behavior data to generate U housing source behavior data blocks of the third template housing source behavior data, wherein the third template housing source behavior data is the i-th candidate template housing source behavior data or the reference template housing source behavior data or the first template housing source behavior data, and U is a positive integer greater than 1;

[0040] Obtaining associated path information of each of the U housing source behavior data blocks in the third template housing source behavior data, encoding and representing the U housing source behavior data blocks, and generating encoding and representing vector data of each of the U housing source behavior data blocks;

[0041] For each of the U housing behavior data blocks, the housing behavior data block is used as a network member of the network knowledge graph, and based on the encoded representation vector data and associated path information of the housing behavior data block, the network member information of the network member is determined to generate the network knowledge graph of the third template housing behavior data.

[0042] 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 housing recommendation processing method based on big data.

[0043] According to a third aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer-executable instructions, and when it is monitored that the computer-executable instructions are executed, the aforementioned housing recommendation processing method based on big data is implemented.

[0044] According to any of the above aspects, the embodiment of the present application realizes efficient and accurate house recommendation by finely processing the big data of house behavior of target users. Specifically, the big data of target house behavior is first decomposed to generate multiple blocks of house behavior data, and then the associated path information of these blocks is obtained and encoded to construct a network knowledge graph containing rich network member information. Based on this network knowledge graph, it is possible to deeply mine the big data of target house behavior and generate a sequence of recommended keywords that reflects user needs, thereby realizing highly personalized house recommendations for target users, significantly improving the accuracy of house recommendations and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] 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.

[0046] Figure 1 A schematic diagram showing a process flow of a housing source recommendation processing method based on big data provided in an embodiment of the present application is shown;

[0047] Figure 2 A schematic diagram of the component structure of a housing recommendation system for implementing the above-mentioned housing recommendation processing method based on big data provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0048] 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.

[0049] 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.

[0050] Figure 1 The flowchart of the method for recommending housing resources based on big data 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 recommending housing resources based on big data can be interchanged according to actual needs, or some steps can be omitted or deleted. The detailed steps of the method for recommending housing resources based on big data are described as follows.

[0051] Step S110, obtaining the target property behavior big data of the target user, and decomposing the target property behavior big data according to the interactive behaviors included in the target property behavior big data to generate X property behavior data blocks of the target property behavior big data.

[0052] For example, the target user Xiao Ming has recently performed a series of house browsing, searching, collecting, consulting and other behaviors on the platform. The data generated by these behaviors constitute the target house behavior big data.

[0053] The server first obtains Xiao Ming’s target property behavior big data. For example, Xiao Ming has performed the following interactive behaviors in the past week:

[0054] 1. Browsed 10 two-bedroom properties in Area A.

[0055] 2. The keyword "three-bedroom apartment near subway station" was searched 5 times.

[0056] 3. Collected 3 properties located in Area B with a total price of less than 2 million.

[0057] 4. Consulted customer service twice for information about school district housing in District C.

[0058] The server decomposes the target property behavior big data based on these interactive behaviors. For example, the browsing behavior data is decomposed into a property behavior data block, the search behavior data is decomposed into a property behavior data block, the favorite behavior data is decomposed into a property behavior data block, and the consultation behavior data is decomposed into a property behavior data block. In this way, 4 property behavior data blocks are generated, that is, X=4.

[0059] Step S120, obtaining the associated path information of each of the X housing source behavior data blocks in the target housing source behavior big data, encoding and representing the X housing source behavior data blocks, and generating encoding representation vector data of each of the X housing source behavior data blocks.

[0060] In this embodiment, for the four house source behavior data blocks generated above, the server obtains the associated path information of each house source behavior data block in the target house source behavior big data.

[0061] Taking the browsing behavior data block as an example, its association path may be that Xiao Ming first searched for "three-bedroom apartment near subway station", and then browsed related listings based on the search results. For the collection behavior data block, the association path may be that after browsing multiple listings, he collected the listings that met certain specific conditions (such as price, location, etc.).

[0062] The server encodes and represents these four property behavior data blocks. For example, a binary encoding method is used to generate a unique code for each data block. The data block encoding of browsing behavior is "0001", the data block encoding of search behavior is "0010", the data block encoding of favorite behavior is "0100", and the data block encoding of consultation behavior is "1000". In this way, the encoding representation vector data of each property behavior data block is generated.

[0063] Step S130, for each of the X housing source behavior data blocks, use the housing source behavior data block as a network member of the network knowledge graph, and determine the network member information of the network member based on the encoded representation vector data and associated path information of the housing source behavior data block, so as to generate the network knowledge graph of the target housing source behavior big data.

[0064] In this embodiment, for each of the four housing source behavior data blocks, the server regards it as a network member of the network knowledge graph.

[0065] Taking the browsing behavior data block as an example, the server determines the network member information of the network member based on its coded representation vector data "0001" and associated path information (such as triggered by search behavior). This may include the member's attributes (such as the type of listings browsed, area, etc.), the relationship with other members (causal relationship with search behavior), etc.

[0066] For the data block of search behavior, its network member information may include search keyword characteristics, search frequency, etc.

[0067] For the data block of collection behavior, the network member information may include the characteristics of the collected properties, the decision factors for collection, etc.

[0068] For the data block of consultation behavior, network member information may include the focus of the consultation issue, the urgency of the consultation, etc.

[0069] By processing each block of listing behavior data, the server finally generates a network knowledge graph of Xiao Ming's target listing behavior big data. This graph clearly shows Xiao Ming's various behaviors in listing-related operations and their interrelationships.

[0070] Step S140, based on the network knowledge graph of the target housing source behavior big data, data mining is performed on the target housing source behavior big data to generate a recommended keyword sequence, and housing sources are recommended to the target user based on the recommended keyword sequence.

[0071] In this embodiment, the server performs data mining on Xiao Ming's target housing behavior big data based on the generated network knowledge graph to generate a recommended keyword sequence.

[0072] First, the server determines a recommendation knowledge base. This recommendation knowledge base contains multiple first template house behavior data, such as:

[0073] 1. Template behavior data 1: Users often search for "sea view room", and the recommended keyword weight distribution is "sea view 0.3, large balcony 0.2, high-end property 0.1".

[0074] 2. Template behavior data 2: The user has collected “small apartment in city center” many times, and the recommended keyword weight distribution is “city center 0.4, small apartment 0.3, convenient transportation 0.2”.

[0075] The server determines the matching score based on the network knowledge graph of Xiao Ming's target housing behavior big data and the knowledge graph vector sequence or network knowledge graph of these first template housing behavior data.

[0076] If Xiao Ming's network knowledge graph shows that he is mainly interested in three-bedroom apartments close to subway stations, which has a high match degree with template behavior data 2, the server will give a higher match score.

[0077] According to the matching score, two (Z=2) template housing source behavior data with a high matching degree with Xiao Ming's behavior are determined from the plurality of first template housing source behavior data. Assume that the determined ones are template behavior data 2 and another template behavior data 3.

[0078] Then, based on the recommended keyword weight distribution of the two template housing behavior data, a recommended keyword sequence is generated. For example, the recommended keyword weight distribution of template behavior data 2 is "city center 0.4, small apartment 0.3, convenient transportation 0.2", and the recommended keyword weight distribution of template behavior data 3 is "low floor 0.3, garden view 0.2, fine decoration 0.2". After comprehensive calculation, the generated recommended keyword sequence may be "city center 0.35, small apartment 0.25, convenient transportation 0.15, low floor 0.1, garden view 0.07, fine decoration 0.08".

[0079] Finally, based on this recommended keyword sequence, the server recommends a house for Xiao Ming. For example, it recommends a small-sized house located in the city center to Xiao Ming, and emphasizes its convenient transportation and other features.

[0080] Assume that this is an operation for a target user group. There are 100 users in the target user group, and each user has big data on house listing behaviors and network knowledge graphs similar to Xiao Ming. Based on the recommended keyword sequences and their weights of each interaction behavior sequence of these 100 users, the server selects an interaction behavior sequence from these 100 interaction behavior sequences as the representative interaction behavior pattern of the target user group.

[0081] For example, after comprehensive analysis, it is found that 80% of users have a high degree of attention and weight on keywords such as "city center" and "convenient transportation". Then the interactive behavior sequence corresponding to this focus and weight distribution can be selected as the representative interactive behavior pattern.

[0082] When determining the recommended knowledge base, the server first obtains a large amount of first template house behavior data, for example, 500 reference template house behavior data and 1,000 candidate template house behavior data.

[0083] For each of the 1000 candidate template listing behavior data, the server determines its matching score with the 500 reference template listing behavior data.

[0084] Taking the 50th candidate template listing behavior data as an example, the server first determines its knowledge graph vector sequence and the knowledge graph vector sequence of each reference template listing behavior data in the 500 reference template listing behavior data. Then, the matching score is determined based on these sequences.

[0085] If the matching score between the 50th candidate template housing source behavior data and at least one reference template housing source behavior data is greater than a second threshold value (eg, 0.7), it is determined as the first template housing source behavior data.

[0086] For the first template housing source behavior data, the server determines its recommended keyword weight distribution based on its matching score with the reference template housing source behavior data.

[0087] Assume that the recommended keyword weight distributions of a first template housing behavior data and multiple reference template housing behavior data with higher matching scores are “good greenery 0.3, large living room 0.2, quiet community 0.2” and “quality property 0.2, ample parking spaces 0.1” respectively, after comprehensive calculation by the server, it is determined that the recommended keyword weight distribution of the first template housing behavior data is “good greenery 0.25, large living room 0.15, quiet community 0.15, quality property 0.2, ample parking spaces 0.1, and others 0.15”.

[0088] When determining the network knowledge graph of the third template housing behavior data, the server also decomposes it according to the interactive behaviors it includes. For example, the third template housing behavior data is that the user compares housing prices in different areas many times, and the server decomposes it into multiple housing behavior data blocks. Then, the associated path information of each block is obtained and encoded, and finally the information of each network member is determined to generate a network knowledge graph.

[0089] Based on the above steps, the embodiment of the present application achieves efficient and accurate house recommendation by finely processing the big data of house behavior of the target user. Specifically, the target house behavior big data is first decomposed to generate multiple house behavior data blocks, and then the associated path information of these blocks is obtained and encoded to construct a network knowledge graph containing rich network member information. Based on this network knowledge graph, it is possible to deeply mine the target house behavior big data and generate a recommended keyword sequence that reflects user needs, thereby achieving highly personalized house recommendations for target users, significantly improving the accuracy of house recommendations and user satisfaction.

[0090] In a possible implementation, step S130 includes:

[0091] Step S131, obtaining behavior pattern statistics of the housing source behavior data block.

[0092] Step S132, determining the network member information of the network member according to the coded representation vector data, associated path information and behavior pattern statistical data of the housing source behavior data block.

[0093] In this embodiment, in step S110, the server has obtained Xiao Li's target property behavior big data and decomposed it into multiple property behavior data blocks, such as browsing behavior data blocks, search behavior data blocks, and consultation behavior data blocks.

[0094] When determining the network member information of the network member: first, the server obtains the behavior pattern statistics of each listing behavior data block.

[0095] Taking the browsing behavior data block as an example, the server counts the frequency with which Xiao Li browses properties within a week, such as an average of 5 times a day; the distribution of browsed property types, such as 60% are two-bedroom properties and 40% are three-bedroom properties; and the regional concentration of browsing, such as 80% of browsing is concentrated in the eastern district of the city.

[0096] For the search behavior data block, the server counts the changing trend of Xiao Li's search keywords, whether they are gradually focused or more divergent; the time distribution of the search, for example, mainly concentrated between 8 and 10 pm.

[0097] For the consultation behavior data block, the server counts the types of consultation questions, such as 40% of inquiries about housing prices, 30% of inquiries about facilities around houses, and 30% of inquiries about property rights; the waiting time for responses to inquiries, etc.

[0098] The server then determines network member information of the network members based on these encoded representation vector data, associated path information, and behavior pattern statistics.

[0099] For the browsing behavior data block, its encoding indicates that the vector data is "0001", and the associated path information shows that it may be triggered by a specific search behavior. The behavior pattern statistics show the browsing frequency, the distribution of housing types, and the regional concentration, etc. Combining this information, the server determines the information of the network member as follows: "The member type is browsing, which is directly related to the search behavior, and the browsing behavior is relatively frequent and concentrated on the two-bedroom and three-bedroom houses in the East District."

[0100] For the search behavior data block, the encoding indicates that the vector data is "0010", and the associated path may point to subsequent browsing and consulting behaviors. The behavior pattern statistics reflect the changes in search keywords and their time distribution. The server determines its network member information as follows: "The member type is search, which triggers subsequent browsing and consulting. The search keywords gradually focus and are mainly active at night."

[0101] For the consultation behavior data block, the encoding represents the vector data as "1000", the associated path is related to browsing and searching behavior, and the behavior pattern statistics present the type of consultation questions and waiting time. The server determines its network member information as: "The member type is consultation, which is associated with browsing and searching behavior. The consultation questions are mainly about housing prices, surrounding facilities and property rights, and the waiting time for response is long."

[0102] By comprehensively considering the encoded representation vector data, associated path information, and behavioral pattern statistics, the server is able to fully and accurately determine the detailed information of each listing behavior data block as a network member, thereby providing a more valuable basis for subsequent analysis and recommendations.

[0103] In a possible implementation, step S140 includes:

[0104] Step S141, determine a recommended knowledge base, the recommended knowledge base includes a recommended keyword weight distribution of each first template property behavior data in Y first template property behavior data, and a knowledge graph vector sequence or a network knowledge graph of each first template property behavior data, the knowledge graph vector sequence of the first template property behavior data is determined based on the network knowledge graph of the first template property behavior data, and Y is a positive integer greater than 1.

[0105] Step S142, determining a matching score between the target property behavior big data and each of the first template property behavior data in the Y first template property behavior data based on the network knowledge graph of the target property behavior big data and the knowledge graph vector sequence or network knowledge graph of each of the first template property behavior data.

[0106] Step S143: Determine Z first template property listing behavior data from the Y first template property listing behavior data based on the matching score, wherein Z is not greater than Y.

[0107] Step S144, generating a recommended keyword sequence according to the recommended keyword weight distribution of the Z first template housing source behavior data.

[0108] In this embodiment, the server first needs to determine a recommended knowledge base. Assume that the recommended knowledge base includes 100 (Y=100) first template house source behavior data.

[0109] For each first template property listing behavior data, the server determines its recommended keyword weight distribution and knowledge graph vector sequence or network knowledge graph.

[0110] For example, the first template housing behavior data is: users frequently search for "garden house", and the recommended keyword weight distribution is "garden 0.4, house 0.3, independent courtyard 0.2, high-end community 0.1", and the corresponding knowledge graph vector sequence is [0.1, 0.2, 0.3, 0.4].

[0111] The second first template housing behavior data is: the user has collected "sea view villa" many times, and the recommended keyword weight distribution is "sea view 0.5, villa 0.3, private swimming pool 0.1, luxury decoration 0.1", and the corresponding knowledge graph vector sequence is [0.2, 0.3, 0.4, 0.1].

[0112] By analogy, the server has determined detailed information for the behavioral data of these 100 first template properties and built a rich recommendation knowledge base.

[0113] Next, the server obtains the target housing behavior big data of a target user (such as Xiao Wang) and generates its network knowledge graph.

[0114] Suppose Xiao Wang's network knowledge graph shows that he is mainly interested in "high-rise apartments in the city center with convenient transportation and complete surrounding facilities."

[0115] The server determines the matching score based on the network knowledge graph of Xiao Wang's target housing behavior big data and the knowledge graph vector sequence or network knowledge graph of the 100 first template housing behavior data.

[0116] For the first template listing behavior data (search for "garden house"), the server uses complex algorithms and models to compare the characteristics and relationships between the two. Since Xiao Wang's focus is quite different from "garden house", the server gives a matching score of 0.2.

[0117] For the second first template property behavior data (collection of "sea view villa"), because Xiao Wang is interested in high-rise apartments in the city center, which are far from sea view villas, the matching score is 0.1.

[0118] In this way, the server compares and calculates the 100 first template housing behavior data one by one, and finally determines the matching score between each first template housing behavior data and Xiao Wang’s target housing behavior big data.

[0119] Based on the matching score calculated previously, the server determines Z data with a higher degree of matching with Xiao Wang's behavior from these 100 first template housing behavior data.

[0120] Assume that the first threshold value is set to 0.6. After screening, there are 5 (Z=5) first template listing behavior data whose matching scores are not less than 0.6.

[0121] The behavior data of these five first template listings are:

[0122] 1. Focus on "small apartments in the city center", with a matching score of 0.7.

[0123] 2. Search for "properties near transportation hubs", the matching score is 0.8.

[0124] 3. Collect “high-rise buildings with complete surrounding facilities”, with a matching score of 0.75.

[0125] 4. Inquiry on "high-quality apartments in the city center", the matching score is 0.65.

[0126] 5. Browse "Properties in Bustling Areas", with a matching score of 0.7.

[0127] **Generate recommended keyword sequence**

[0128] The server generates a recommended keyword sequence based on the recommended keyword weight distribution of the five first template property behavior data.

[0129] The recommended keyword weight distribution of the first template housing behavior data (focusing on "small apartments in the city center") is "city center 0.4, small apartment 0.3, good lighting 0.2, low property fee 0.1".

[0130] The recommended keyword weight distribution of the second first template housing behavior data (search for "housing near transportation hubs") is "transportation hub 0.5, nearby housing 0.3, convenient travel 0.1, developed business 0.1".

[0131] The recommended keyword weight distribution of the third first template housing behavior data (collection of "high-rise buildings with complete surrounding supporting facilities") is "surrounding supporting facilities 0.4, high-rise buildings 0.3, good greening 0.2, and sufficient parking spaces 0.1".

[0132] The recommended keyword weight distribution of the fourth first template housing behavior data (inquiry on "high-quality apartments in the city center") is "city center 0.4, high-quality apartments 0.3, property services 0.2, security system 0.1".

[0133] The recommended keyword weight distribution of the fifth first template housing behavior data (browsing "real estate in a bustling area") is "bustling location 0.4, real estate 0.3, appreciation potential 0.2, brand developer 0.1".

[0134] The server comprehensively considers the weight distribution of these five recommended keywords and performs weighted calculations. For example, "city center" appears twice, with weights of 0.4 and 0.4 respectively, and the weighted average is 0.4. The weighted averages of all keywords are calculated in this way, and the final recommended keyword sequence may be: "city center 0.4, small apartment 0.25, transportation hub 0.15, surrounding facilities 0.1, high-rise 0.05, good lighting 0.05, convenient travel 0.05, good greening 0.03, sufficient parking space 0.02, property service 0.02, security system 0.01, appreciation potential 0.01, brand developer 0.01".

[0135] Based on this recommended keyword sequence, the server can accurately recommend properties that meet Xiao Wang's needs, such as showing him a small apartment located in the city center, focusing on its transportation convenience, surrounding supporting facilities, house lighting and other features.

[0136] Assume that this is not for a single user, but a target user group. There are 500 users in this user group, and each user has his or her own target property behavior big data and corresponding network knowledge graph.

[0137] The server also follows the above steps to determine a matching score for each user, screens out Z first template property behavior data with a higher matching degree, and generates a recommended keyword sequence.

[0138] Then, the server combined the recommended keyword sequences and their weights of these 500 users for further analysis and mining. For example, it was found that 80% of users included keywords such as "city center" and "convenient transportation" in their recommended keyword sequences, and their weights were relatively high.

[0139] Based on such analysis, the server can select a representative interaction behavior pattern from the interaction behavior sequences of these 500 users, such as an interaction behavior sequence that mainly focuses on "small-sized high-rise apartments in the city center, with convenient transportation and complete surrounding facilities" as the representative interaction behavior pattern of this target user group.

[0140] In this way, the recommendation knowledge base and target housing behavior big data can be effectively utilized to generate accurate and useful recommendation keyword sequences for users or user groups, and provide personalized housing recommendation services.

[0141] In a possible implementation, step S142 includes:

[0142] Step S1421, if the recommended knowledge base includes the network knowledge graph of the Y first template property behavior data, then determine the matching score between the target property behavior big data and each of the first template property behavior data in the Y first template property behavior data based on the network knowledge graph of the target property behavior big data and the network knowledge graph of the Y first template property behavior data.

[0143] Step S1422, if the recommended knowledge base includes the knowledge graph vector sequence of the Y first template property behavior data, then determine the matching score between the target property behavior big data and each of the first template property behavior data in the Y first template property behavior data based on the knowledge graph vector sequence of the target property behavior big data and the knowledge graph vector sequence of the Y first template property behavior data.

[0144] First, the server has built a recommendation knowledge base, which contains 200 (Y=200) first template listing behavior data.

[0145] Case 1: The recommended knowledge base contains the network knowledge graph of the first template housing behavior data: the server obtains the target housing behavior big data of the target user Xiao Li and generates its network knowledge graph.

[0146] For example, Xiao Li’s online knowledge graph shows that he frequently searches for “three-bedroom apartments near schools” and browses in multiple areas. He also has a specific interest in the orientation and price of the houses.

[0147] For the network knowledge graph of the first template housing behavior data, it shows that users mainly search for "sea view villas, focusing on landscape and area". The server compares Xiao Li's network knowledge graph with the network knowledge graph of the first template housing behavior data and finds that the two are quite different in terms of housing type and focus, so it gives a lower matching score, such as 0.2.

[0148] For the network knowledge graph of the second first template housing behavior data, it shows that the user searches for "one-bedroom apartment near the city center, focusing on transportation and surrounding facilities". After comparison, the server finds that although both focus on location and surrounding facilities, the housing types and demand sizes are different, and gives a medium matching score, such as 0.5.

[0149] In this way, the server performs detailed feature comparison and relationship analysis on the network knowledge graphs of the 200 first template housing behavior data with Xiao Li's network knowledge graph one by one, so as to determine the matching score between each first template housing behavior data and Xiao Li's target housing behavior big data.

[0150] Case 2: The recommended knowledge base contains the knowledge graph vector sequence of the first template housing behavior data: the server obtains the target housing behavior big data of the target user Xiao Zhang and generates its knowledge graph vector sequence.

[0151] For example, Xiao Zhang’s knowledge graph vector sequence is [0.3, 0.2, 0.4, 0.1], which indicates his attention level and behavioral characteristics in certain aspects.

[0152] For the first template housing behavior data, the knowledge graph vector sequence is [0.2, 0.3, 0.1, 0.4]. The server calculates the similarity or difference between the two through a specific algorithm, such as calculating the cosine value of the angle between the vectors. It is found that the similarity is low and a matching score of 0.3 is given.

[0153] For the second first template housing behavior data, the knowledge graph vector sequence is [0.3, 0.2, 0.3, 0.2]. After calculation, the server finds that it has a certain similarity with Xiao Zhang and gives a matching score of 0.6.

[0154] The server processes the knowledge graph vector sequences of the 200 first template property behavior data and Xiao Zhang's knowledge graph vector sequence in this calculation method to determine the matching score between each first template property behavior data and Xiao Zhang's target property behavior big data.

[0155] Through detailed processing of the above two situations, the matching degree between the target users and the behavioral data of numerous first template listings can be accurately determined, providing strong support for subsequent recommendations and analysis.

[0156] In a possible implementation, step S1421 includes:

[0157] Step S1421-1, determining the knowledge graph vector sequence of the target property source behavior big data based on the network knowledge graph of the target property source behavior big data.

[0158] Step S1421-2: Determine the knowledge graph vector sequence of the Y first template housing source behavior data based on the network knowledge graph of the Y first template housing source behavior data.

[0159] Step S1421-3, based on the knowledge graph vector sequence of the target property behavior big data and the knowledge graph vector sequence of the Y first template property behavior data, determine the matching score between the target property behavior big data and each of the first template property behavior data in the Y first template property behavior data.

[0160] First, the server obtains the big data of the target user Xiao Zhao's target housing behavior and generates his network knowledge graph. For example, Xiao Zhao's network knowledge graph shows that he frequently browses four-bedroom houses in the new urban area, pays special attention to the greening and parking spaces in the community, and has repeatedly consulted about school districts.

[0161] The server determines the knowledge graph vector sequence based on the network knowledge graph of Xiao Zhao's target property behavior big data through specific algorithms and models. Assume that this vector sequence is [0.3, 0.4, 0.2, 0.1], which respectively represents Xiao Zhao's attention level or behavior characteristics in different aspects.

[0162] At the same time, the server already has 500 (Y=500) network knowledge graphs of the first template housing behavior data. For the first network knowledge graph of the first template housing behavior data, it shows that users are mainly interested in two-bedroom apartments in the old city, and are more concerned about the decoration style of the house and the surrounding commercial facilities. Based on this, the server determines that the knowledge graph vector sequence of this first template housing behavior data is [0.2, 0.3, 0.4, 0.1].

[0163] For the network knowledge graph of the second first template housing behavior data, it shows that users focus on searching for three-bedroom apartments close to parks, and pay attention to the orientation of the house and property fees. The server determines that its knowledge graph vector sequence is [0.3, 0.2, 0.1, 0.4].

[0164] By analogy, the server determines the corresponding knowledge graph vector sequence for the network knowledge graph of the 500 first template housing behavior data.

[0165] Next, the server determines the matching score based on the knowledge graph vector sequence [0.3, 0.4, 0.2, 0.1] of Xiao Zhao's target property behavior big data and the knowledge graph vector sequence of the 500 first template property behavior data.

[0166] For example, for the first knowledge graph vector sequence [0.2, 0.3, 0.4, 0.1] of the first template listing behavior data, the server determines the matching score by calculating the similarity or distance between the two vector sequences. Assuming that the cosine similarity calculation method is used, the calculated similarity is 0.5, and the server determines it as the matching score of the two.

[0167] For the knowledge graph vector sequence [0.3, 0.2, 0.1, 0.4] of the second first template housing behavior data, the calculated similarity is 0.3, which the server uses as the matching score.

[0168] In this way, the server compares and calculates the knowledge graph vector sequences of the 500 first template property behavior data with the knowledge graph vector sequences of Xiao Zhao's target property behavior big data one by one, and finally determines the matching score between each first template property behavior data and Xiao Zhao's target property behavior big data.

[0169] In this way, the similarity and matching between the target user's housing behavior data and each first template housing behavior data can be accurately measured, providing a strong basis for subsequent housing recommendations and user behavior analysis.

[0170] In a possible implementation, step S143 includes:

[0171] From the Y first template housing source behavior data, determine Z first template housing source behavior data whose matching score with the target housing source behavior big data is not less than a first threshold value.

[0172] In a possible implementation, step S144 includes:

[0173] If Z is greater than 1, the average recommended keyword weight distribution of the recommended keyword weight distributions of the Z first template house source behavior data is output as the recommended keyword weight distribution of the target house source behavior big data.

[0174] In this embodiment, in the previous steps, the server has calculated the matching scores between the target housing source behavior big data of the target user Xiao Wang and 300 (Y=300) first template housing source behavior data.

[0175] First, set the first threshold value to 0.7. The server starts to screen from these 300 matching scores. For example, the matching score of the first first template listing behavior data is 0.6, which does not meet the requirements; the matching score of the second first template listing behavior data is 0.8, which meets the requirements; the matching score of the third first template listing behavior data is 0.75, which meets the requirements.

[0176] After screening one by one, the server finally determined 50 (Z=50) first template property behavior data with a matching score of no less than 0.7.

[0177] Next, the server obtains the recommended keyword weight distribution of the 50 first template house source behavior data. Assume that the recommended keyword weight distribution of one of the first template house source behavior data is "convenient transportation 0.3, surrounding facilities 0.2, house area 0.2, decoration style 0.1, property management 0.1, other 0.1", and the recommended keyword weight distribution of another first template house source behavior data is "school district house 0.4, house orientation 0.2, reasonable price 0.2, community environment 0.1, developer brand 0.1", and so on.

[0178] Since Z(50) is greater than 1, the server calculates the average value of the weight distribution of these 50 recommended keywords.

[0179] Taking the keyword "convenient transportation" as an example, the server adds up the weight values ​​of "convenient transportation" in the behavior data of the 50 first template properties, and then divides it by 50 to obtain its weight value in the average recommended keyword weight distribution.

[0180] In this way, the server calculates each keyword and finally obtains the recommended keyword weight distribution of the target housing behavior big data, such as "convenient transportation 0.25, surrounding facilities 0.18, house area 0.15, decoration style 0.12, property management 0.1, school district housing 0.08, house orientation 0.06, reasonable price 0.05, community environment 0.03, developer brand 0.02, and others 0.01".

[0181] In a possible implementation, if the target property behavior big data is the i-th interaction behavior sequence among C key property interaction behavior sequences of the target user group, C is a positive integer, and i is a positive integer less than or equal to C, the method further includes: based on the recommended keyword sequence and its weight of each interaction behavior sequence in the C key property interaction behavior sequences, selecting an interaction behavior sequence from the C key property interaction behavior sequences as a representative interaction behavior pattern of the target user group.

[0182] Assume that the server is processing a target user group on a real estate platform, which contains 100 users (C=100), each of whom has his or her key property interaction behavior sequence.

[0183] For example, the recommended keyword sequence and its weight for the first user (i.e., the first interactive behavior sequence) are: “low total price 0.4, small apartment 0.3, close to subway 0.2, fine decoration 0.1”.

[0184] The recommended keyword sequence and its weights for the second user (i.e., the second interactive behavior sequence) are: “school district housing 0.5, three-bedroom apartment 0.3, good greening 0.1, complete supporting facilities 0.1”.

[0185] And so on, until the 100th user.

[0186] The server first performs a detailed analysis of the recommended keyword sequences and their weights of these 100 key property interaction behavior sequences.

[0187] The server found that among the many recommended keywords, the keyword "low total price" appeared in the recommended keyword sequences of multiple users and had a high weight. Through further statistics and calculations, the server found that 80 users had "low total price" in their recommended keyword sequences, and their average weight reached 0.35.

[0188] At the same time, the keyword "near subway" has also attracted the attention of many users. It is included in the recommended keyword sequence of 70 users, with an average weight of 0.25.

[0189] Taking into account the frequency of occurrence and weight distribution of each keyword, the server finally determines the interaction behavior sequence of the 30th user as the representative interaction behavior pattern of the target user group. The recommended keyword sequence and its weight for this user are: "low total price 0.3, near subway 0.25, small apartment 0.2, complete supporting facilities 0.15, convenient transportation 0.1".

[0190] In this way, a representative pattern can be determined from the interaction behavior sequences of many users in order to better understand and meet the common needs of the entire target user group.

[0191] In a possible implementation, step S141 includes:

[0192] Step S1411, obtaining the Y first template housing behavior data.

[0193] Step S1412, determining the network knowledge graph and recommended keyword weight distribution of each first template property behavior data in the Y first template property behavior data.

[0194] Step S1413, determining the recommended knowledge base based on the network knowledge graph of the Y first template housing source behavior data and the weight distribution of recommended keywords.

[0195] For example, in a possible implementation, step S1411 includes:

[0196] Obtain S reference template housing source behavior data and T candidate template housing source behavior data, where both S and T are positive integers. Based on the S reference template housing source behavior data, select at least one from the T candidate template housing source behavior data as the first template housing source behavior data.

[0197] In this embodiment, first, the server needs to obtain Y first template house source behavior data. In this process, the server first obtains 200 (S=200) reference template house source behavior data and 500 (T=500) candidate template house source behavior data.

[0198] The 200 reference templates cover a variety of user behavior patterns, such as:

[0199] -Reference template 1: Users frequently search for "high-rise apartments, focusing on landscape views", and the recommended keyword weight distribution is "high-rise 0.4, landscape view 0.3, floor-to-ceiling windows 0.2, high-quality properties 0.1". The network knowledge graph shows that their attention to house height and view is closely related.

[0200] -Reference template 2: The user has collected "single-family villa, pursuit of private space" many times, and the recommended keyword weight distribution is "single-family 0.5, private space 0.3, large garden 0.1, independent garage 0.1". The network knowledge graph shows his pursuit of independence and space of the house.

[0201] For the 500 candidate template listing behavior data, the server screens them based on the 200 reference template listing behavior data.

[0202] For example, for the 100th candidate template listing behavior data, the server determines its matching score with the 200 reference template listing behavior data. By comparing and calculating its behavior characteristics, focus points, etc., it is found that its matching score with the 50th reference template listing behavior data is relatively high, reaching 0.8 (assuming the set threshold value is 0.7). Because it exceeds the threshold value, the server determines this 100th candidate template listing behavior data as the first template listing behavior data.

[0203] In this way, the server processes and screens the behavior data of 500 candidate template properties one by one, and finally determines the behavior data of 300 first template properties.

[0204] Then, the server determines the network knowledge graph and recommended keyword weight distribution of each of the 300 first template housing behavior data.

[0205] For example, for one of the first template housing behavior data, its network knowledge graph shows that users pay more attention to the location of the house and the surrounding schools during the search and browsing process, and there is a strong correlation between the two. The weight distribution of its recommended keywords is "high-quality school district 0.4, downtown location 0.3, convenient transportation 0.2, and living facilities 0.1".

[0206] The server performs such processing and determination on the behavior data of the 300 first template properties.

[0207] Finally, based on the network knowledge graph of the behavior data of the 300 first template properties and the weight distribution of recommended keywords, the server successfully determined the recommended knowledge base. In the subsequent user behavior data processing and property recommendation, the server can use this recommended knowledge base to provide users with more accurate and more in line with their needs services.

[0208] For example, in a possible implementation, the selecting at least one of the S reference template housing source behavior data from the T candidate template housing source behavior data as the first template housing source behavior data includes:

[0209] For the i-th candidate template property source behavior data among the T candidate template property source behavior data, determine the matching score between the i-th candidate template property source behavior data and each reference template property source behavior data among the S reference template property source behavior data, where i is a positive integer not greater than T.

[0210] If the matching score between the i-th candidate template housing source behavior data and at least one of the S reference template housing source behavior data is greater than a second threshold value, the i-th candidate template housing source behavior data is output as the first template housing source behavior data.

[0211] In this embodiment, the server has obtained 100 (S=100) reference template house source behavior data and 300 (T=300) candidate template house source behavior data.

[0212] For the 50th (i=50) candidate template listing behavior data among the 300 candidate template listing behavior data, the server begins to determine its matching score with each of the 100 reference template listing behavior data.

[0213] Assume that the behavior data of the 50th candidate template listing shows that users are mainly interested in "sea view room, balcony, fine decoration". The server first compares it with the behavior data of the first reference template listing, which is "mountain view room, large living room, simple decoration". By analyzing the differences between the two in terms of listing characteristics and user concerns, the server calculates a matching score of 0.2.

[0214] Next, the server compares the behavior data of the 50th candidate template house with the behavior data of the 2nd reference template house, which is "sea view room, floor-to-ceiling windows, luxurious decoration". Since both are concerned about sea view rooms and have certain similarities in decoration, the server calculates a matching score of 0.7.

[0215] In this way, the server compares the behavior data of the 50th candidate template listing with the behavior data of 100 reference template listings in turn and calculates the matching score.

[0216] Assume that the second threshold value is set to 0.6. Since the matching score 0.7 between the 50th candidate template house source behavior data and the second reference template house source behavior data is greater than the second threshold value 0.6, the server outputs the 50th candidate template house source behavior data as the first template house source behavior data.

[0217] The server follows the same process to perform such processing on each of the 300 candidate template property behavior data, thereby screening out the first template property behavior data that meets the conditions for subsequent operations such as recommendation knowledge base construction and user behavior analysis.

[0218] For example, in a possible implementation, determining a matching score between the i-th candidate template housing source behavior data and each reference template housing source behavior data in the S reference template housing source behavior data includes:

[0219] Determine the knowledge graph vector sequence of the i-th candidate template property behavior data and the knowledge graph vector sequence of each reference template property behavior data in the S reference template property behavior data.

[0220] Based on the knowledge graph vector sequence of the i-th candidate template property behavior data and the knowledge graph vector sequence of each reference template property behavior data in the S reference template property behavior data, determine the matching score between the i-th candidate template property behavior data and each reference template property behavior data in the S reference template property behavior data.

[0221] For example, in a possible implementation, determining the knowledge graph vector sequence of the i-th candidate template property behavior data and the knowledge graph vector sequence of each reference template property behavior data in the S reference template property behavior data includes:

[0222] Determine the network knowledge graph of the i-th candidate template housing source behavior data and the network knowledge graph of each reference template housing source behavior data in the S reference template housing source behavior data.

[0223] Based on the network knowledge graph of the i-th candidate template housing source behavior data, determine the knowledge graph vector sequence of the i-th candidate template housing source behavior data.

[0224] For each reference template housing source behavior data in the S reference template housing source behavior data, a knowledge graph vector sequence of the reference template housing source behavior data is determined based on the network knowledge graph of the reference template housing source behavior data.

[0225] In this embodiment, it is assumed that the server is processing the template house source behavior data of the real estate platform, and the 88th (i=88) candidate template house source behavior data is taken as an example.

[0226] First, the server determines the network knowledge graph of the 88th candidate template housing behavior data. Assume that this network knowledge graph shows that users frequently search for "duplex near the park, focusing on lighting and ventilation".

[0227] At the same time, the server also determines the network knowledge graph of each data in 50 (S=50) reference template housing behavior data.

[0228] For example, the network knowledge graph of the first reference template housing behavior data shows "a bungalow near a school, focusing on area and layout"; the network knowledge graph of the second reference template housing behavior data shows "a high-rise near a shopping mall, focusing on floor level and view", and so on.

[0229] Next, the server determines the knowledge graph vector sequence of the 88th candidate template property behavior data through a specific algorithm and model. Assume that the vector sequence is [0.3, 0.4, 0.2, 0.1], which respectively represents the degree of attention to factors such as parks, duplexes, lighting, and ventilation.

[0230] For each of the 50 reference template property behavior data, the server also determines the knowledge graph vector sequence based on its network knowledge graph.

[0231] For example, the knowledge graph vector sequence of the first reference template housing behavior data is [0.4, 0.3, 0.2, 0.1], the knowledge graph vector sequence of the second reference template housing behavior data is [0.2, 0.3, 0.4, 0.1], and so on.

[0232] After obtaining the knowledge graph vector sequence of the 88th candidate template property behavior data and the knowledge graph vector sequence of the 50 reference template property behavior data, the server determines the matching score by calculating indicators such as the similarity or distance between them.

[0233] Assume that for the first reference template property behavior data, the calculated matching score is 0.2; for the second reference template property behavior data, the calculated matching score is 0.5.

[0234] In this way, the server calculates the matching score between the 88th candidate template property behavior data and each of the 50 reference template property behavior data.

[0235] In this way, the degree of match between the behavior data of the 88th candidate template property and the behavior data of each reference template property can be accurately determined, providing strong data support for subsequent screening and analysis.

[0236] For example, in a possible implementation, determining the recommended keyword weight distribution of each of the Y first template housing source behavior data includes:

[0237] If the first template property source behavior data is the i-th candidate template property source behavior data, then based on the matching score between the i-th candidate template property source behavior data and each reference template property source behavior data in the S reference template property source behavior data, T target template property source behavior data whose matching score with the i-th candidate template property source behavior data is not less than a third threshold value are determined from the S reference template property source behavior data.

[0238] According to the recommended keyword weight distributions of the T candidate template housing source behavior data, the recommended keyword weight distribution of the i-th candidate template housing source behavior data is determined.

[0239] In this embodiment, it is assumed that the server has determined 80 (Y=80) first template housing source behavior data, wherein the 30th first template housing source behavior data is screened out from the candidate template housing source behavior data (ie, the 30th candidate template housing source behavior data).

[0240] The server obtains the matching score between the 30th candidate template listing behavior data and each of the 200 (S=200) reference template listing behavior data. Assume that the third threshold value is set to 0.7.

[0241] After comparison, the server finds that there are 50 reference template property behavior data whose matching score with the 30th candidate template property behavior data is not less than 0.7 (T=50).

[0242] The recommended keyword weight distribution of the 50 target template property behavior data is as follows:

[0243] Target template 1: "Convenient transportation 0.4, surrounding facilities 0.3, house area 0.2, decoration style 0.1"

[0244] Target template 2: "School district housing 0.5, house orientation 0.2, reasonable price 0.2, community environment 0.1"

[0245] …

[0246] Target template 50: "Low floors 0.3, garden landscape 0.3, property management 0.2, parking space 0.2"

[0247] The server conducts a comprehensive analysis of the recommended keyword weight distribution of the 50 target template property behavior data.

[0248] For example, for the keyword "convenient transportation", the server calculates the weighted average of "convenient transportation" in the behavior data of the 50 target template listings. Assume that the calculated result is 0.35.

[0249] For other keywords, such as "surrounding facilities", "school district housing", "housing area", etc., the server calculates the average in the same way.

[0250] Finally, the server determined that the recommended keyword weight distribution of the 30th candidate template housing behavior data (i.e. the 30th first template housing behavior data) was "convenient transportation 0.35, surrounding facilities 0.25, school district housing 0.15, house area 0.1, house orientation 0.05, reasonable price 0.03, community environment 0.03, low floor 0.02, garden landscape 0.01, property management 0.01, parking space 0.01".

[0251] Therefore, the weight distribution of recommended keywords is accurately determined for each first template housing behavior data, so as to provide users with more accurate housing recommendation services in the future.

[0252] For example, in a possible implementation, determining a network knowledge graph of the third template housing behavior data includes:

[0253] According to the interactive behaviors included in the third template house source behavior data, the third template house source behavior data is decomposed to generate U house source behavior data blocks of the third template house source behavior data, the third template house source behavior data is the i-th candidate template house source behavior data or the reference template house source behavior data or the first template house source behavior data, and U is a positive integer greater than 1.

[0254] Obtain associated path information of each of the U housing source behavior data blocks in the third template housing source behavior data, encode and represent the U housing source behavior data blocks, and generate encoding representation vector data of each of the U housing source behavior data blocks.

[0255] For each of the U housing behavior data blocks, the housing behavior data block is used as a network member of the network knowledge graph, and based on the encoded representation vector data and associated path information of the housing behavior data block, the network member information of the network member is determined to generate the network knowledge graph of the third template housing behavior data.

[0256] In this embodiment, the following is an example of a detailed scenario performed with the server as the execution subject:

[0257] Assume that the server is processing a third template listing behavior data, which is the 20th candidate template listing behavior data.

[0258] First, the server decomposes the interactive behaviors included in the 20th candidate template listing behavior data. Assuming that this data includes the user's search, browse, favorite, and consult four interactive behaviors, the server will generate 4 (U=4) listing behavior data blocks.

[0259] Next, the server obtains the associated path information of these four property behavior data blocks in the 20th candidate template property behavior data. For example, the server finds that the user first searches for "low total price properties", then browses several related properties based on the search results, then collects one of the properties that meets the budget, and finally asks the customer service for detailed information about the property.

[0260] Afterwards, the server encodes and represents these four property behavior data blocks. Assuming the hexadecimal encoding method is used, the data block encoding of the search behavior is "0001", the data block encoding of the browsing behavior is "0002", the data block encoding of the collection behavior is "0004", and the data block encoding of the consultation behavior is "0008", thereby generating the encoding representation vector data of each property behavior data block.

[0261] Finally, for each of these four property behavior data blocks, the server treats it as a network member of the network knowledge graph.

[0262] For the data block of the search behavior (coded "0001"), the server determines the network member information of the network member based on its coded representation vector data and associated path information (triggered by the user's initial demand). For example, its attribute is determined as "the initial demand is a low total price", and its relationship with other members is "triggered subsequent browsing, collection and consultation behaviors".

[0263] For the data block of browsing behavior (coded “0002”), the network member information may be “viewed multiple listings based on the search results, and paid attention to the detailed features of the listings.”

[0264] For the data block of collection behavior (coded “0004”), the network member information may be “selected the properties that met specific conditions to collect from the browsed properties, indicating a strong interest”.

[0265] For the data block of consulting behavior (coded “0008”), the network member information may be “want to learn more about the favorite properties and is interested in answers to specific questions.”

[0266] By processing each block of listing behavior data, the server finally generates a network knowledge graph of the 20th candidate template listing behavior data. This graph clearly shows the various behaviors of users in this candidate template listing behavior and their interrelationships.

[0267] In the same steps, if the third template property listing behavior data is a reference template property listing behavior data or the first template property listing behavior data, the server will also process it in the above manner to generate a corresponding network knowledge graph.

[0268] 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.

[0269] 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, to implement the above-mentioned method for processing house recommendation based on big data. 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 examples, 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.

[0270] The transmission device 106 is used to obtain or send data via a fault diagnosis 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 fault diagnosis 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.

[0271] 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.

[0272] 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.

[0273] 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 method for processing house recommendation based on big data, characterized in that: The method comprises: Obtaining target housing behavior big data of a target user, and decomposing the target housing behavior big data according to the interactive behaviors included in the target housing behavior big data to generate X housing behavior data blocks of the target housing behavior big data; Obtaining associated path information of each of the X housing source behavior data blocks in the target housing source behavior big data, encoding and representing the X housing source behavior data blocks, and generating encoding and representing vector data of each of the X housing source behavior data blocks; For each of the X housing source behavior data blocks, the housing source behavior data block is used as a network member of the network knowledge graph, and network member information of the network member is determined according to the encoding representation vector data and the associated path information of the housing source behavior data block, so as to generate the network knowledge graph of the target housing source behavior big data; According to the network knowledge graph of the target housing source behavior big data, data mining is performed on the target housing source behavior big data to generate a recommendation keyword sequence, and housing sources are recommended to the target user based on the recommendation keyword sequence; The method of performing data mining on the target property behavior big data based on the network knowledge graph of the target property behavior big data to generate a recommended keyword sequence includes: Determine a recommended knowledge base, the recommended knowledge base comprising a recommended keyword weight distribution of each first template property behavior data in Y first template property behavior data, and a knowledge graph vector sequence or a network knowledge graph of each first template property behavior data, wherein the knowledge graph vector sequence of the first template property behavior data is determined based on the network knowledge graph of the first template property behavior data, and Y is a positive integer greater than 1; Determine a matching score between the target property behavior big data and each of the first template property behavior data according to the network knowledge graph of the target property behavior big data and the knowledge graph vector sequence or network knowledge graph of each of the first template property behavior data; Determine Z first template property listing behavior data from the Y first template property listing behavior data according to the matching score, where Z is not greater than Y; Generate a recommended keyword sequence according to the recommended keyword weight distribution of the Z first template housing source behavior data.

2. The method for processing house recommendation based on big data according to claim 1, characterized in that: The determining of the network member information of the network member according to the coded representation vector data and the associated path information of the housing source behavior data block includes: Obtaining behavior pattern statistics of the housing source behavior data block; The network member information of the network members is determined based on the coded representation vector data, associated path information and behavior pattern statistical data of the housing source behavior data block.

3. The method for processing house recommendation based on big data according to claim 1, characterized in that: The step of determining a matching score between the target property behavior big data and each of the first template property behavior data based on the network knowledge graph of the target property behavior big data and the knowledge graph vector sequence or network knowledge graph of each of the first template property behavior data includes: If the recommendation knowledge base includes the network knowledge graphs of the Y first template property behavior data, then determining a matching score between the target property behavior big data and each of the first template property behavior data in the Y first template property behavior data based on the network knowledge graph of the target property behavior big data and the network knowledge graph of the Y first template property behavior data; If the recommended knowledge base includes the knowledge graph vector sequence of the Y first template property behavior data, the matching score between the target property behavior big data and each of the first template property behavior data in the Y first template property behavior data is determined based on the knowledge graph vector sequence of the target property behavior big data and the knowledge graph vector sequence of the Y first template property behavior data.

4. The method for processing house recommendation based on big data according to claim 3, characterized in that: The determining, based on the network knowledge graph of the target housing source behavior big data and the network knowledge graph of the Y first template housing source behavior data, a matching score between the target housing source behavior big data and each first template housing source behavior data in the Y first template housing source behavior data comprises: Determine a knowledge graph vector sequence of the target property source behavior big data based on the network knowledge graph of the target property source behavior big data; Determine a knowledge graph vector sequence of the Y first template housing source behavior data according to the network knowledge graph of the Y first template housing source behavior data; Based on the knowledge graph vector sequence of the target property behavior big data and the knowledge graph vector sequence of the Y first template property behavior data, a matching score between the target property behavior big data and each of the Y first template property behavior data is determined.

5. The method for processing house recommendation based on big data according to claim 1, characterized in that: Determining Z first template housing source behavior data from the Y first template housing source behavior data based on the matching score includes: From the Y first template housing source behavior data, determine Z first template housing source behavior data whose matching score with the target housing source behavior big data is not less than a first threshold value.

6. The method for processing house recommendation based on big data according to claim 1, characterized in that: The generating a recommended keyword sequence according to the recommended keyword weight distribution of the Z first template housing source behavior data includes: If Z is greater than 1, the average recommended keyword weight distribution of the recommended keyword weight distributions of the Z first template house source behavior data is output as the recommended keyword weight distribution of the target house source behavior big data.

7. The method for processing house recommendation based on big data according to claim 1, characterized in that: If the target housing source behavior big data is the i-th interaction behavior sequence among C key housing source interaction behavior sequences of the target user group, C is a positive integer, and i is a positive integer less than or equal to C, the method further includes: Based on the recommended keyword sequence and its weight of each interaction behavior sequence in the C key property interaction behavior sequences, an interaction behavior sequence is selected from the C key property interaction behavior sequences as a representative interaction behavior pattern of the target user group.

8. The method for processing house recommendation based on big data according to claim 1, characterized in that: The determining of the recommended knowledge base comprises: Obtaining the Y first template house listing behavior data; Determine a network knowledge graph and a recommended keyword weight distribution of each first template housing source behavior data in the Y first template housing source behavior data; The recommended knowledge base is determined based on the network knowledge graph of the Y first template housing source behavior data and the weight distribution of recommended keywords.

9. 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 housing recommendation processing method based on big data described in any one of claims 1 to 8 is implemented.

Images