An AI intelligent house source search matching method of an apartment rental operation system
By constructing explicit and implicit demand vectors and using an AI matching model to match apartment rental listings, the problem of insufficient understanding of user needs in existing technologies is solved, resulting in more accurate listing recommendations and personalized displays.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZHEJIANG YUNBEI INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2025-11-17
- Publication Date
- 2026-06-16
AI Technical Summary
The existing apartment rental system cannot deeply understand user needs, resulting in inaccurate matching results and difficulty in providing personalized recommendations.
By collecting users' search and behavioral data, explicit and implicit demand vectors are constructed. An AI matching model is used to generate a list of matching properties, and property attribute vectors are combined to perform property matching.
It improves the accuracy of property matching and the effectiveness of personalized recommendations, thus optimizing the user experience.
Smart Images
Figure CN121478836B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data matching technology, specifically to an AI-powered intelligent property search and matching method for an apartment rental operation system. Background Technology
[0002] With rapid urbanization and a large influx of people into cities, the demand for apartment rentals is growing rapidly. However, traditional apartment rental models rely heavily on offline agents, personal recommendations, or keyword searches on rental platforms, requiring users to manually filter listings that match their criteria, resulting in low efficiency. Meanwhile, landlords and operators also hope to more efficiently recommend suitable properties to potential tenants, improving rental efficiency and property utilization. This can be achieved by adopting intelligent management systems that not only update property status in real time but also optimize rental pricing strategies through data analysis, attracting more high-quality tenants and maximizing resource utilization.
[0003] Currently, apartment rental information services mainly rely on conditional filtering or collaborative filtering algorithms for property matching. These algorithms can filter properties based on keywords or preset conditions entered by users on rental platforms and recommend potentially interesting properties by analyzing user browsing, favorites, rental history, and the preferences of similar users. However, filtering conditions are usually based on simplistic criteria such as price, size, and location, lacking an understanding and exploration of users' deeper needs. The needs expressed by users during searches are often complex and vague, but the system can only process keywords or limited conditions. It fails to fully utilize multi-dimensional features such as user search text data and user behavior data for comprehensive analysis, resulting in inaccurate matching results and difficulty in providing effective personalized recommendations. Summary of the Invention
[0004] To address the technical problem that existing apartment rental systems cannot deeply understand and analyze users, thus affecting the accuracy of matching results and hindering personalized recommendations, the present invention aims to provide an AI-powered intelligent property search and matching method for apartment rental operation systems. The specific technical solution adopted is as follows:
[0005] The system collects user search and behavioral data through the apartment rental operation system.
[0006] Based on search data, the explicit demand vector of users for renting a house is determined by fusing current search data and historical search data respectively;
[0007] Analyze behavioral data to determine user behavior patterns and attention to housing listings, and identify the implicit demand vector of users for renting housing;
[0008] An AI matching model is established, which constructs a property attribute vector through the apartment rental operation system, and combines explicit and implicit demand vectors to generate a list of matching properties using the AI matching model.
[0009] Preferably, the search data includes search time, user's fixed preferences set according to fixed options in the apartment rental operation system, and input search text; the behavioral data includes behavioral time, browsing behavior, interaction behavior, and historical tenant data.
[0010] Preferably, determining the explicit demand vector for users' rental needs includes:
[0011] Based on predefined tags for fixed options, clean the search text of the current search data and perform word segmentation. Convert each word segment into a word vector and use tags to label fixed preferences and word segmentation to construct an explicit demand vector for the current search data.
[0012] Analyze any historical search data, determine the reference weight, and merge it with the explicit demand vector of the current search to determine the explicit demand vector of the user's rental housing.
[0013] Preferably, any historical search data is analyzed to determine the reference weight, specifically as follows:
[0014] Based on historical search data, we determine the reference characteristics of user demand behavior, and determine the demand stability index by analyzing the word segmentation of historical search data. We also calculate the search time difference between historical search data and current search data to obtain the reference value characteristics of historical search data, and then normalize them to obtain reference weights.
[0015] Preferably, the reference value characteristics of historical search data include:
[0016] Based on historical search data, user interest and user scrolling characteristics are determined to obtain reference characteristics of user demand behavior.
[0017] Obtain the similarity between any word in the current historical search data and every word in other historical search data, and sort and construct a similarity sequence for each word. Determine the similar demand words for all words through the similarity sequence, and obtain all similarities with the current word. Combine the weighted average of user demand behavior reference features to obtain the repetition feature index of all words, and determine the demand stability index.
[0018] By statistically analyzing the search time difference between historical search data and current search data, we can obtain the reference value characteristics of historical search data.
[0019] Preferably, determining the implicit demand vector of users renting a house includes:
[0020] The time and duration of each user's access to the apartment rental operation system are obtained, and user behavior time characteristic parameters and housing interest area behavior characteristics are obtained respectively.
[0021] The confidence scores for user behavior time characteristics and interest area characteristics are obtained by analyzing the time of each user's access to the apartment rental operation system and the behavioral characteristics of the housing interest areas.
[0022] Based on behavioral time feature parameters, behavioral features of housing interest areas, confidence of behavioral time features and confidence of interest areas, the implicit demand intensity and implicit demand confidence are determined, and a user's implicit demand vector for renting a house is constructed.
[0023] Preferably, the user's behavioral time characteristic parameters and the behavioral characteristics of the house interest area are obtained, including:
[0024] To obtain sunrise time, the feature time is obtained by subtracting the time of each user's access to the apartment rental operation system from the previous adjacent sunrise time. The user's behavioral time feature parameters are then obtained through the feature time.
[0025] The information displayed on each property details page in the apartment rental operation system is defined as the area of interest. The dwell time in the area of interest is counted and clustered to obtain multiple clusters.
[0026] The high-frequency interest areas are identified by filtering the clusters, and the user's house interest area behavior characteristics are obtained based on the high-frequency interest areas.
[0027] Preferably, the confidence scores for the user's behavioral time features and the confidence scores for the region of interest are obtained, including:
[0028] The standard deviation of the time interval is determined by the time of each user's access to the apartment rental operation system, and the confidence level of the user's behavioral time characteristics is obtained.
[0029] Based on the time spent in high-frequency attention areas, the confidence level of users' attention areas is obtained by combining the time spent in attention areas.
[0030] Preferably, the implicit demand intensity includes the ratio of behavioral time feature parameters to the confidence level of behavioral time features, and the product of behavioral features of the house's area of interest and the confidence level of the area of interest; the implicit demand confidence level includes the ratio of behavioral time feature confidence level to the confidence level of the area of interest.
[0031] Preferably, generating a list of matching properties includes:
[0032] By comparing explicit demand vectors with property attribute vectors and implicit demand vectors with property attribute vectors using an AI matching model, the explicit constraint matching score and correlation degree are obtained.
[0033] The system combines explicit constraint matching scores and relevance scores to output a comprehensive score for each property, and then sorts these scores to generate a list of matched properties.
[0034] The present invention has the following beneficial effects:
[0035] The system analyzes collected user search and behavioral data. It integrates current and historical search data to construct an explicit demand vector, unifying the dimensions of the search data and creating a structured, computable, and interpretable explicit demand feature vector. Then, based on behavioral data, it determines user behavior patterns and property interest, uncovering implicit needs and identifying an implicit demand vector. Finally, it employs an AI (Artificial Intelligence) matching model combined with property attribute vectors to generate a list of matching properties. In short, by working together with property attribute vectors, explicit and implicit demand vectors, and utilizing deep learning models, the system improves the accuracy of matching user needs with properties, optimizes the display of properties based on user needs, enhances personalized property matching, increases the accuracy of personalized recommendations, and improves the user experience. Attached Figure Description
[0036] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is a flowchart illustrating the steps of an AI-powered intelligent property search and matching method for an apartment rental operation system, as provided in an embodiment of the present invention. Detailed Implementation
[0038] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an AI-powered intelligent housing search and matching method for an apartment rental operation system proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0040] The following description, in conjunction with the accompanying drawings, details the specific scheme of the AI-powered intelligent housing search and matching method for an apartment rental operation system provided by this invention.
[0041] Existing apartment operation systems only filter and match based on simple conditions, while user needs are usually complex and vague. They cannot understand and explore users' deeper needs, nor can they fully utilize the multidimensional features of needs, resulting in low matching accuracy and difficulty in providing personalized recommendations. Therefore, this paper proposes to analyze collected user search and behavioral data to capture users' explicit needs and uncover their implicit needs, and construct corresponding explicit and implicit demand vectors for users' rental needs. Then, an AI matching model is used in conjunction with property attribute vectors to generate a list of matched properties, thereby improving the accuracy of property matching and displaying property information that is closer to users' needs.
[0042] Please see Figure 1 The diagram illustrates a flowchart of the steps in an AI-powered intelligent housing search and matching method for an apartment rental operation system provided in the first embodiment of the present invention. The method includes:
[0043] Step S1: Collect user search data and behavioral data through the apartment rental operation system;
[0044] Step S2: Based on the search data, determine the explicit demand vector of users for renting a house by fusing the current search data and historical search data respectively;
[0045] Step S3: Analyze behavioral data to determine user behavior patterns and attention to housing listings, and identify the implicit demand vector of users for renting housing;
[0046] Step S4: Establish an AI matching model. Construct property attribute vectors through the apartment rental operation system, and combine explicit and implicit demand vectors to generate a list of matching properties using the AI matching model.
[0047] As an optional implementation method, the apartment rental operation system targets various types of housing resources, including apartments, residential housing, and commercial properties. Among them, commercial properties include offices, shops, or warehouses to meet the needs of different enterprises and individuals. Preferably, in the actual application of the method, the analysis is based on the apartment rental operation system and its database. Its housing resources, transportation, and other information can be directly used in the subsequent processing without special acquisition.
[0048] To better illustrate, the user's rental process is typically a dynamic, phased process encompassing three stages: First, the exploration stage, usually involving a general browsing and casual examination of various listings; second, the convergence stage, where users typically add filters and compare details of each listing, including photos, descriptions, and reviews; and third, the decision-making stage, involving scheduling viewings, conducting on-site inspections, and face-to-face communication with landlords for desired properties, ultimately placing an order after considering all factors. The behavioral implications and preference confidence levels differ at each stage. In-depth analysis and matching of user needs requires comprehensive data, including basic property information such as size, layout, floor, orientation, and decoration style; user search and behavioral data; market dynamics such as housing price trends and policy changes; and environmental factors such as surrounding transportation, safety, and noise levels. This comprehensive data foundation, usable by algorithms, provides data support for subsequent property-user matching.
[0049] Furthermore, search data includes search time, user preferences set by users in the apartment rental operation system based on fixed options, and the search text entered; behavioral data includes behavior time, browsing behavior, interaction behavior, and historical tenant data.
[0050] It can be explained that search time refers to time-related data such as the start time or duration of each user's search; the fixed options in the apartment rental operation system usually cover rent range, location preference, room type, pet-friendly, etc.; and the apartment rental operation system is equipped with a search box to record the search text entered by the user in real time, which serves as a supplementary column for requirements, allowing users to describe their personalized preferences such as not wanting to be on the street or wanting it to be quiet at night in short sentences.
[0051] Behavioral time refers to the time spent accessing search-related data, browsing property listings, scheduling viewings, etc.; browsing behavior includes the number of times a user browses property listings, zooming in on image areas, etc.; interactive behavior includes user actions such as saving, sharing, scheduling viewings, etc.; historical tenant data refers to historical tenant records, including the user's historical rental duration, number of rentals, and other related rental history.
[0052] Understandably, analyzing search and behavioral data can help determine a user's true rental intentions. However, since the data is presented in different dimensions and forms, in order for the apartment rental operation system to better understand the relevant data and build a comprehensive model, the relevant data can be converted into a user demand feature vector. This vector can be seen as a mathematical representation of user preferences, allowing for in-depth mining of user rental needs. This includes not only explicit needs reflected in rigid conditions such as price and size, but also implicit needs incorporating fuzzy needs, contextual needs, and behavioral cues. This provides a data foundation for the subsequent matching process of housing that meets user needs.
[0053] Further, in step S2, the explicit demand vector for users' rental needs is determined, including:
[0054] Step S21: Based on the predefined labels of the fixed options, clean the search text of the current search data and perform word segmentation. Convert each word segment into a word vector and use labels to annotate fixed preferences and word segmentation to construct an explicit demand vector for the current search data.
[0055] It can be explained that the explicit demand vector of users renting a house is captured based on search data. That is, the fixed options selected by users in the apartment rental operation system and the search text entered are uniformly abstracted into a structured, computable and interpretable explicit demand feature vector, which is used as the explicit rental demand directly expressed by users.
[0056] Specifically, predefined tags include budget, location, commute, pets, room type, desired attributes, and unwanted attributes, which are used to map explicit needs. In particular, the predefined tags contain at least all fixed preference options that users can filter, that is, predefined tags for all fixed options to ensure that tags can be matched according to the user's filtering.
[0057] Next, the search text entered by the user is cleaned to remove invisible characters, emojis, etc. The cleaned search text includes the current search text and the adjacent previous historical search text. In reality, users usually select their ideal property after multiple searches, so it is necessary to perform vertical analysis on the search data to ensure the accuracy of the current search data analysis.
[0058] Next, the cleaned search text was segmented using the word segmentation tool jieba. This open-source Chinese word segmentation tool based on the Python language decomposes the text into independent words, obtaining several Chinese word segmentation information. Then, the BERT (Bidirectional Encoder Representations from Transformers) model was used to convert each word segmentation into a vector representation, which yields several word vectors for subsequent data analysis.
[0059] Finally, each word segment is labeled using the BiLSTM-CRF (Bidirectional Long Short-Term Memory - Conditional Random Field) sequence labeling model. This model utilizes the bidirectional long short-term memory network to capture contextual information in the search text, while combining it with a conditional random field for global optimization to ensure the accuracy and consistency of labeling. Selected fixed preferences are also labeled, constructing a multi-dimensional vector, denoted as the explicit demand vector of the current search data. Each label corresponds to one dimension, and when no word segment or fixed preference is assigned in the predefined labels, a uniform null value is assigned.
[0060] Step S22: Analyze any historical search data, determine the reference weight, and fuse it with the explicit demand vector of the current search to determine the explicit demand vector of the user's rental housing.
[0061] Understandably, in the process of auxiliary analysis of historical search data, different historical search data have varying reference value for users' actual needs. Some historical search data can reflect users' stable preferences well. For example, if a user stays on a search result page for a long time and clicks or delves into multiple property details, it indicates that the search is more likely to match the user's actual needs and has higher reference value. On the other hand, some historical search data may lack representativeness due to user dissatisfaction with search results, biased input, or erroneous operations, and cannot accurately reflect the user's actual needs. For example, if a user stays on a search result page for a short time, lacks in-depth clicking, or even quickly scrolls through pages, it may mean that the search results do not meet the user's expectations, or that the search was just a random attempt, and its reference value should be reduced. If search data is not differentiated, noise can easily be introduced into the construction of the demand vector, reducing the accuracy of judging users' actual needs. Therefore, analyzing historical search data and determining corresponding reference weights can improve the accuracy and effectiveness of explicit demand vectors.
[0062] Further, in step S22, any historical search data is analyzed to determine the reference weight, specifically as follows:
[0063] Based on historical search data, we determine the reference characteristics of user demand behavior, and determine the demand stability index by analyzing the word segmentation of historical search data. We also calculate the search time difference between historical search data and current search data to obtain the reference value characteristics of historical search data, and then normalize them to obtain reference weights.
[0064] It can be explained that historical search data can be used as an important indicator of whether a search can consistently represent user intent, through the repetitive characteristics of the text. Specifically, if certain word segments appear repeatedly as core keywords in multiple searches, it indicates that the conditions corresponding to these word segments are highly stable in the user's rental needs, thus possessing greater reference value and warranting higher weighting. Furthermore, users' rental needs are typically dynamic during the process of selecting rental properties. The closer the search time of historical search data is to the current search time, the more it reflects the user's latest needs. Earlier searches are more likely to be outdated and should have lower weighting.
[0065] For better illustration, in this embodiment, the historical search data analyzed is selected from the user's Ath search data.
[0066] Furthermore, the reference value characteristics of historical search data are obtained, including:
[0067] Step S221: Determine user interest and user swipe characteristics based on historical search data to obtain user demand behavior reference characteristics.
[0068] Specifically, the total browsing time for users to click on various property listings on the search results page obtained from the Ath search data is used as the user interest level for the Ath search data, denoted as . The average display time of all property listings on the search results page obtained from the A-th search is used as the user scrolling feature of the A-th search data, denoted as . Based on the ratio calculated from user interest and user swipe characteristics, a reference feature for user demand behavior is obtained, denoted as . ,Right now This represents the user demand behavior reference feature of the Ath search data. The larger the value, the higher the user's interest in and the more detailed the browsing of the search results in the Ath search data, and the more valuable it is for reference. Similarly, the user demand behavior reference feature of each historical search data can be obtained.
[0069] To better understand the average display time, let's assume a user is browsing on a mobile phone. Different property listings are displayed in different areas of the screen. Assuming the search results page can display 6 property listings, and the user scrolls through the page until the next listing appears (the 7th listing), if the previous listing (the 1st listing) disappears in 10 seconds, this includes 2 seconds for the next listing to appear. Therefore, the 1st listing will only disappear from the current search results page while the 7th listing is fully displayed. The display time for the 7th listing is 2 seconds, and the display time for the 1st listing stops being calculated. The display times for listings 2 through 6 continue to be calculated. Assuming the user exits the search results page after the 7th listing fully appears, the average display time for all property listings in this search is: .
[0070] Step S222: Obtain the similarity between any word in the current historical search data and every word in other historical search data, and sort and construct a similarity sequence for each word. Determine the similar demand words of all words through the similarity sequence, and obtain all similarities with the current word. Combine the weighted average of user demand behavior reference features to obtain the repetition feature index of all words, and determine the demand stability index.
[0071] Specifically, based on step S21, all word segments in the A-th search data and other historical search data are obtained, and their corresponding word vector representations are obtained. The cosine similarity of the word vectors of each word segment in the A-th search data and each word segment in other historical search data is calculated. That is, assuming the word segments in the A-th search data are... Other historical search data, that is, assuming the word segmentation in the Bth search data is... , , All represent the total number of word segments; the word segments being analyzed are: Then calculate in sequence. and , and , , and The word vector cosine similarity is used to assess the closeness between two word segments in the semantic space; then, word segments are established by sorting them in descending order based on the word vector cosine similarity. The similarity sequence is obtained; similarly, the similarity sequence of each word in the Ath search data is obtained.
[0072] Then, based on the word segmentation similarity sequence of any word segmentation, all elements corresponding to the larger of the two elements with the largest difference between adjacent elements are segmented, and these are determined as the word segmentation for the current word segmentation similarity requirement. For better understanding, we still focus on word segmentation. The analysis is performed, assuming that the constructed similarity sequence is... In this similarity sequence, the differences between adjacent elements are 0.17, 0.32, 0.12, and 0.18, respectively. The maximum difference between adjacent elements is 0.32. Comparing based on 0.32, the differences between 0.73 and 0.56, 0.56 and 0.44, and 0.44 and 0.32 are all greater than or equal to 0.32. The corresponding pairs of elements meet the requirements, and the larger elements are 0.73, 0.56, and 0.44, respectively. The similarity of these three elements is based on the current word segmentation. The words were calculated by comparing them with the corresponding words in other historical search data. Therefore, the corresponding words of the three elements in other historical search data were used as the current words. The similarity requirement word segmentation is obtained; similarly, the similarity requirement word segmentation of all words in the Ath search data is obtained.
[0073] Next, based on the similarity of the segmented words with the current segment, all similarities are obtained. Using the user demand behavior reference features corresponding to the search data of each similar demand segment, the similarity weight between the similar demand segment and the current segment is calculated and amplified. Then, softmax normalization is used, and the weighted average is calculated to obtain the repetition feature index of the current segment, denoted as [index]. This improves the accuracy of the word segmentation repetition feature index in reflecting user needs; similarly, the repetition feature index of all words in the A-th search data is obtained, and the index is averaged to obtain the demand stability index, denoted as... .
[0074] Step S223: Calculate the search time difference between historical search data and current search data to obtain the reference value characteristics of historical search data.
[0075] Specifically, in this embodiment, the search time difference between historical search data (i.e., the A-th search data) and the current search data is calculated, normalized, and denoted as . The reference value characteristics of historical search data are obtained, that is, the reference value characteristics of the A-th search data in the current analysis are obtained. The corresponding calculation formula is:
[0076]
[0077] in, This indicates the reference value characteristics of the Ath search data being analyzed in the current analysis; This represents the user demand behavior reference characteristics of the Ath search data; This represents the demand stability index of the Ath search data. This represents the search time difference between the A-th search data after normalization and the current search data. To correct the parameter, this embodiment of the invention sets it to 0.1 to prevent the denominator from being 0. It can be adjusted according to the specific implementation environment.
[0078] Similarly, the reference value features of each historical search data are obtained, and softmax normalization is performed to obtain the reference weight of each historical search data.
[0079] It can be explained that in step S22, the explicit demand vector of the user's rental needs is determined by fusing the reference weight of each historical search data with the explicit demand vector of the current search, so as to obtain a more stable explicit demand vector, which can improve the accuracy and noise resistance of the explicit demand vector understanding.
[0080] The explicit demand vector of the current search data, where each tag corresponds to a dimension, and when there are no assigned words or fixed preferences in the predefined tags, a uniform null value is assigned.
[0081] Specifically, non-null dimension data in the explicit demand vector of the current search data are retained. Optionally, in this embodiment, a priority weight is set. The current search data for any non-null dimension is fused with historical search data using priority weights. The weight of the current search data for any non-null dimension is 0.7, while the weight of the historical search data is used as a reference weight. A weighted average is then applied to obtain the comprehensive value for each non-null dimension, which serves as the non-null dimension data corresponding to the user's explicit demand vector for renting. For null dimensions in the explicit demand vector of the current search data, a weighted fusion method is used with the corresponding dimensions from historical search data to obtain the null dimension data corresponding to the user's explicit demand vector. Specifically, if there are no non-null dimension data in the historical search data, that dimension is not included in the fusion analysis to ensure the reliability of the dimension data. This results in a more stable and representative explicit demand vector from the current search data, which serves as the user's explicit demand vector for renting, denoted as [vector name missing]. .
[0082] Understandably, during the interaction process based on search data, user behavior data is analyzed to obtain the implicit needs of users renting houses. That is, implicit needs are extracted from users' browsing, filtering, interaction behavior and historical rental records, and solidified into a user implicit need vector to improve the accuracy and noise resistance of housing matching. Implicit needs refer to needs that users do not express directly in the search text, but whose behavioral data clearly shows their preferences.
[0083] To better illustrate, behavioral data includes behavioral time, which refers to the time when users access the apartment rental operation system. Users' choice to access the apartment rental operation system to search during specific time periods is often closely related to their lifestyle habits rather than being random. It can reflect the user's life rhythm and potential needs and preferences. In other words, corresponding analysis can determine the current user's behavioral patterns. For example, users who frequently access and search late at night may have higher requirements for the safety of nighttime travel and living environment due to working late. Thus, this time behavior pattern can be regarded as an implicit characteristic of the user's lifestyle, providing supplementary information for judging user needs.
[0084] Secondly, browsing behavior on property details pages can reveal users' attentional biases and implicit expectations, reflecting their focus on each property listing. When users frequently zoom in on different living areas such as the kitchen and bedroom, or repeatedly read descriptions of amenities, it indicates that these features are actually important to them, even if not explicitly stated in their filter criteria. Meanwhile, some needs may not be explicitly expressed because users assume they are necessary conditions, such as ample natural light, good ventilation, or basic security. Therefore, by analyzing users' focus on details pages, these passive, implicit needs can be further identified.
[0085] Further, in step S3, the implicit demand vector for users' rental needs is determined, including:
[0086] Step S31: Obtain the time and duration of each user's access to the apartment rental operation system, and obtain the user's behavioral time characteristic parameters and the behavioral characteristics of the housing interest area.
[0087] To clarify, the time a user visits the apartment rental operation system refers to the time when the user initially enters the system during a search, usually the moment the page loads after the user clicks the login button or enters the URL; the duration corresponds to the length of time it takes to obtain search data, that is, the time spent from starting the search to finally obtaining all relevant rental information.
[0088] Further, in step S31, the user's behavioral time characteristic parameters and the behavioral characteristics of the house interest area are obtained, including:
[0089] Step S311: Obtain the sunrise time. Based on the difference between the time of each user's access to the apartment rental operation system and the adjacent previous sunrise time, obtain the feature time. Obtain the user's behavioral time feature parameters through the feature time.
[0090] As an alternative implementation, sunrise time is obtained via the Internet and varies dynamically based on the seasons or weather changes.
[0091] Specifically, characteristic time refers to a specific time period used to reflect a user's lifestyle and potential needs and preferences. It is the difference between the time a user visits the apartment rental operation system each time and the previous sunrise time. For example, if a user visits the apartment rental operation system at 6:00 AM and the previous sunrise time was 5:30 AM, the characteristic time is 30 minutes, indicating that the user's active period is in the early morning.
[0092] The behavioral time feature parameter refers to the weighted average time of a user's visits to the apartment rental operation system. The weight is represented by the duration of each user's visit to the system after softmax normalization, amplifying the weight of the user's long-term visits to search for rental information. This results in the current user's behavioral time feature parameter, denoted as... The larger the value, the greater the probability that the user is a late-night worker. This indicates that the user is more likely to have a greater need for a safe living environment and convenient nighttime transportation. In other words, they are more inclined to choose residential areas with good lighting and security measures. They also have higher requirements for whether there are 24-hour public transportation or taxi services nearby. In addition, for the sake of convenience, they may also pay attention to whether there are 24-hour convenience stores, pharmacies and other living facilities in their current residential area to ensure that their daily needs can be met conveniently in the late night or early morning.
[0093] Step S312: Define the information displayed on each property details page in the apartment rental operation system as the area of interest, count the dwell time in the area of interest and cluster them to obtain multiple clusters.
[0094] The explanation provided indicates that the areas of focus include images of the living room, bedroom, or kitchen displayed on the property details page, as well as textual information such as area, room type, floor, orientation, amenities, and surrounding environment.
[0095] Specifically, the time users spend in the areas of interest on all property listings is counted, and clustering is performed to obtain multiple clusters. That is, users' preferences and behavioral patterns for areas of interest are identified based on the time spent in the cluster. After clustering, areas of interest with similar time spent in the cluster are grouped together to form a cluster.
[0096] Step S313: Filter clusters to determine high-frequency interest areas, and obtain the user's house interest area behavior characteristics based on the high-frequency interest areas.
[0097] Specifically, the average dwell time is calculated for each cluster, and all areas of interest in the cluster with the highest average dwell time are selected as high-frequency areas of interest. The time a user spends viewing each high-frequency area of interest is calculated, and the difference between this time and the time a user spends accessing the apartment rental operation system is calculated. A weighted average is then calculated based on this difference. The weight is the ratio of the dwell time corresponding to a high-frequency area of interest to the total time spent accessing the apartment rental operation system, after softmax normalization. This weight amplifies the evaluation proportion of information that users frequently focus on (i.e., high-frequency areas of interest) and that has a shorter access time to the apartment rental operation system, thus inferring the housing information corresponding to shorter dwell times in areas of interest, thereby improving the accuracy of the interest feature evaluation for each area. The weighted average is then used as the user's housing interest area behavior feature, denoted as […]. The larger this value, the more obvious the user's attention is to the high-frequency areas corresponding to the property.
[0098] Step S32: Obtain the confidence scores of the user's behavioral time characteristics and the confidence scores of the areas of interest by using the time and area of interest of each user's access to the apartment rental operation system.
[0099] The explanation is as follows: the confidence score of behavioral time characteristics is used to reflect the behavioral patterns of users in relation to the apartment rental operation system; the confidence score of attention area is used to reflect the user's attention to the housing listings.
[0100] Further, in step S32, the confidence scores of the user's behavioral time features and the confidence scores of the region of interest are obtained, including:
[0101] Step S321: Determine the standard deviation of the time interval by the time of each user's access to the apartment rental operation system, and obtain the confidence level of the user's behavioral time characteristics.
[0102] Specifically, based on the time of all visits to the apartment rental operation system, the time interval between two adjacent visits is calculated, and the average is calculated and the square root is taken to obtain the standard deviation of the time interval. The range of the standard deviation is then set using the tanh function. The confidence score of the user's behavioral time feature is then denoted as... The smaller the value, the stronger the regularity of the user's current rental behavior, and the more reliable the implicit needs reflected in the user's behavioral time characteristics.
[0103] Step S322: Based on the dwell time statistics of high-frequency attention areas, the user's confidence level of attention areas is obtained by combining the dwell time of attention areas.
[0104] Specifically, the time users spend viewing frequently viewed areas of each property is calculated, along with the total time users spend on each property. The ratio of these two times is obtained. Similarly, the ratios for all frequently viewed areas are calculated, and the average is calculated. The tanh function is then used to set the range of this average value to... This is then used as the confidence level of the user's region of interest, denoted as... The larger this value, the more reliable the implicit needs reflected in the user's focus on the housing area.
[0105] Step S33: Based on behavioral time feature parameters, behavioral features of the housing interest area, confidence of behavioral time features and confidence of interest area, determine the implicit demand intensity and implicit demand confidence, and construct the implicit demand vector of users renting houses.
[0106] It can be explained that the implicit demand vector is also composed of several dimensions, each dimension representing a possible type of implicit demand, such as nighttime safety demand, nighttime transportation convenience demand, kitchen functionality demand, bedroom functionality demand, etc. Each dimension contains two parts: implicit demand intensity and implicit demand confidence. Implicit demand intensity includes two parts: behavioral characteristics and housing concerns. Implicit demand confidence is explained by combining the two parts.
[0107] Furthermore, in step S33, the implicit demand intensity includes the ratio of the behavioral time characteristic parameter to the confidence level of the behavioral time characteristic, and the product of the behavioral characteristics of the housing interest area and the confidence level of the interest area; the implicit demand confidence level includes the ratio of the confidence level of the behavioral time characteristic to the confidence level of the interest area.
[0108] Preferably, the explanation is based on the Xth dimension of the implicit demand vector, where the intensity of the implicit demand is: and The normalized value, assuming the Xth dimension represents nighttime security needs, is the intensity of implicit needs. Corresponding behavioral time analysis, behavioral time characteristic parameters The larger the confidence level of the behavioral time features, the greater the confidence level. The smaller the value, the higher the probability that the user is a late-night worker, and the stronger the regularity of their rental behavior, indicating a higher demand for safety in their living environment; while the intensity of implicit needs... Analyzing behavior related to housing inquiries reveals that behavioral data changes with each search, indicating the intensity of implicit demand. Based on behavioral data, these characteristics will also change accordingly. For example, let's assume that a user's behavioral characteristics regarding the area of interest in a house are as follows in the current behavioral data. and confidence level of the area of interest The larger the values, the more attention users pay to the information presented on the property details page, and because they are likely to return late, they are also more concerned about security.
[0109] Next, the ratio of the calculated behavioral time feature confidence score to the attention area confidence score is used as the implicit demand confidence score for the Xth dimension. Similarly, the two components of the implicit demand intensity for each dimension and the implicit demand confidence score are determined, and then integrated to obtain the implicit demand vector for users renting a house, denoted as... .
[0110] Understandably, further analysis is conducted based on the explicit and implicit needs of users seeking rental housing, generating a list of search-matched properties for users, optimizing the display effect, and ensuring that each recommended property meets the user's personalized preferences and quality of life requirements.
[0111] It can be explained that in step S4, a housing attribute vector is constructed through the apartment rental operation system, wherein the housing attribute vector is denoted as... It is built based on the database of the apartment rental operation system itself, including basic housing attributes such as location, rent, apartment type, and area; traffic information such as the density and distance of surrounding subway and bus lines; community environmental data such as access control and security information; and the coverage of surrounding streetlights.
[0112] Further, in step S4, a list of matching properties is generated, including:
[0113] Step S41: Using the AI matching model, compare the explicit demand vector with the property attribute vector and the implicit demand vector with the property attribute vector to obtain the explicit constraint matching score and correlation degree.
[0114] As an optional implementation, the AI matching model adopts a deep ranking model with a multi-tower structure, including an explicit demand matching tower, an implicit demand preference tower, and an interaction fusion layer.
[0115] Specifically, the explicit demand matching tower uses a fully connected neural network to match explicit demand vectors. With property attribute vector By performing a dimension-by-dimensional comparison and analyzing the data differences in each dimension, the system accurately captures the fit between the user's explicit needs and the characteristics of the property, and outputs an explicit constraint matching score. This score ensures that candidate properties strictly meet the user's basic screening criteria, avoids recommendation results that do not meet the hard criteria, and can recommend relevant properties more accurately, thereby improving user experience and satisfaction.
[0116] The latent demand preference pyramid uses an attention-enhanced neural network to capture the latent demand vector. With property attribute vector The relevance of a property is determined by using a multi-layered attention mechanism to uncover property features that users do not explicitly express but are highly interested in, thereby recommending property information that better matches their needs.
[0117] Step S42: Combine the explicit constraint matching score and relevance to output the comprehensive score of each property, and sort them to generate a list of matched properties.
[0118] It can be explained that the interactive fusion layer outputs a comprehensive score for each property. The interactive fusion layer uses a multi-gate mixed-of-experts (MMOE) or deep interest network (DIN) to integrate the outputs of the explicit demand matching tower and the implicit demand preference tower. It has an attention mechanism that dynamically adjusts the fusion weights of different dimensions based on the implicit demand confidence. That is, when the implicit demand confidence is higher, the contribution of that dimension in the implicit demand vector to the final ranking is enhanced, thereby improving the personalization level of the recommended properties and the user experience.
[0119] Specifically, the AI matching model outputs a comprehensive score for each property, i.e., f(explicit constraint matching score + relevance) = comprehensive score. Here, the explicit constraint matching score serves as a hard constraint, ensuring that candidate properties, i.e., recommended properties, meet the user's basic conditions. The relevance, as an optimization factor, further distinguishes and personalizes properties that meet the constraints, following the logic of "meeting first, then optimizing." Next, each candidate property is sorted from high to low based on the comprehensive score, forming a list of matched properties. This list not only ensures the effectiveness of the recommended properties but also improves the fit between the recommended properties and the user's potential interests, thereby increasing the efficiency of the user's property selection process.
[0120] Understandably, based on the analysis of collected user search and behavioral data, the system integrates current and historical search data to construct an explicit demand vector. This unifies the dimensions of the search data, creating a structured, computable, and interpretable explicit demand feature vector. Then, based on behavioral data, it determines user behavior patterns and property interest levels, uncovering implicit needs and identifying an implicit demand vector. Finally, an AI (Artificial Intelligence) matching model is used in conjunction with property attribute vectors to generate a list of matching properties. In short, through the synergy between property attribute vectors, explicit and implicit demand vectors, and deep learning models, the system improves the accuracy of matching user needs with properties, optimizes the display of properties based on user needs, enhances the effectiveness of personalized property matching, strengthens the accuracy of personalized recommendations, and improves the user experience.
[0121] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0122] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
Claims
1. An AI-powered intelligent property search and matching method for an apartment rental operation system, characterized in that, The method includes: The system collects user search and behavioral data through the apartment rental operation system. Based on search data, the explicit demand vector of users for renting a house is determined by fusing current search data and historical search data respectively; Analyze behavioral data to determine user behavior patterns and attention to housing listings, and identify the implicit demand vector of users for renting housing; Establish an AI matching model, construct property attribute vectors through the apartment rental operation system, and generate a list of matching properties by combining explicit and implicit demand vectors using the AI matching model; The process of determining the explicit demand vector for users' rental needs includes: Based on predefined tags for fixed options, clean the search text of the current search data and perform word segmentation. Convert each word segment into a word vector and use tags to label fixed preferences and word segmentation to construct an explicit demand vector for the current search data. Analyze any historical search data, determine the reference weight, and fuse it with the explicit demand vector of the current search to determine the explicit demand vector of the user's rental housing. The process of constructing an explicit demand vector for the current search data by using tags to label fixed preferences and word segmentation includes: Each word segment is labeled using the BiLSTM-CRF sequence labeling model. This involves using a bidirectional long short-term memory network to capture contextual information in the search text and combining it with a conditional random field for global optimization to ensure the accuracy and consistency of labeling. Selected fixed preferences are also labeled, constructing a multi-dimensional vector, denoted as the explicit demand vector of the current search data. Each label corresponds to one dimension, and when no word segment or fixed preference is assigned in the predefined labels, a uniform null value is assigned.
2. The AI-powered intelligent housing search and matching method for an apartment rental operation system according to claim 1, characterized in that, The search data includes search time, user preferences set by the user in the apartment rental operation system based on fixed options, and the search text entered; the behavioral data includes behavioral time, browsing behavior, interaction behavior, and historical tenant data.
3. The AI-powered intelligent housing search and matching method for an apartment rental operation system according to claim 2, characterized in that, Analyze any historical search data to determine the reference weight, specifically: Based on historical search data, we determine the reference characteristics of user demand behavior, and determine the demand stability index by analyzing the word segmentation of historical search data. We also calculate the search time difference between historical search data and current search data to obtain the reference value characteristics of historical search data, and normalize them to obtain reference weights. The process of obtaining the user demand behavior reference features includes: Based on historical search data, user interest and user scrolling characteristics are determined to obtain reference characteristics of user demand behavior; the user scrolling characteristic is the average display time of all housing information displayed on the search results page for each search. The process of obtaining the demand stability index includes: Obtain the similarity between any word in the current historical search data and every word in other historical search data, and sort them to construct a similarity sequence for each word. Determine the similar demand words for all words through the similarity sequence, and obtain all similarities with the current word. Combine the user demand behavior reference features and calculate the weighted average to obtain the repetition feature index of all words. Average the repetition feature index of all words to obtain the demand stability index.
4. The AI-powered intelligent housing search and matching method for an apartment rental operation system according to claim 2, characterized in that, Identify the implicit demand vector of users renting housing, including: The time and duration of each user's access to the apartment rental operation system are obtained, and user behavior time characteristic parameters and housing interest area behavior characteristics are obtained respectively. The confidence scores for user behavior time characteristics and interest area characteristics are obtained by analyzing the time of each user's access to the apartment rental operation system and the behavioral characteristics of the housing interest areas. Based on behavioral time feature parameters, behavioral features of housing interest areas, confidence of behavioral time features and confidence of interest areas, the implicit demand intensity and implicit demand confidence are determined, and a user's implicit demand vector for renting a house is constructed.
5. The AI-powered intelligent housing search and matching method for an apartment rental operation system according to claim 4, characterized in that, Obtain user behavior time characteristic parameters and house interest area behavior characteristics, including: To obtain sunrise time, the feature time is obtained by subtracting the time of each user's access to the apartment rental operation system from the previous adjacent sunrise time. The user's behavioral time feature parameters are then obtained through the feature time. The information displayed on each property details page in the apartment rental operation system is defined as the area of interest. The dwell time in the area of interest is counted and clustered to obtain multiple clusters. The high-frequency interest areas are identified by filtering the clusters, and the user's house interest area behavior characteristics are obtained based on the high-frequency interest areas.
6. The AI-powered intelligent housing search and matching method for an apartment rental operation system according to claim 5, characterized in that, Obtain the confidence scores for the user's behavioral time characteristics and the confidence scores for the region of interest, including: The standard deviation of the time interval is determined by the time of each user's access to the apartment rental operation system, and the confidence level of the user's behavioral time characteristics is obtained. Based on the time spent in high-frequency attention areas, the confidence level of users' attention areas is obtained by combining the time spent in attention areas.
7. The AI-powered intelligent housing search and matching method for an apartment rental operation system according to claim 4, characterized in that, The implicit demand intensity includes the ratio of behavioral time feature parameters to the confidence level of behavioral time features, and the product of behavioral features of the housing interest area and the confidence level of the interest area; the implicit demand confidence level includes the ratio of behavioral time feature confidence level to the confidence level of the interest area.
8. The AI-powered intelligent housing search and matching method for an apartment rental operation system according to claim 1, characterized in that, Generate a list of matching properties, including: By comparing explicit demand vectors with property attribute vectors and implicit demand vectors with property attribute vectors using an AI matching model, the explicit constraint matching score and correlation degree are obtained. The system combines explicit constraint matching scores and relevance scores to output a comprehensive score for each property, and then sorts these scores to generate a list of matched properties.