A residential replacement matching system based on information analysis

Abstract

The present application relates to a kind of based on information analysis's residential replacement matching system, including picture feature processing subsystem, user demand analysis module, replacement matching index module and matching preference screening module;The system generates concrete house features with house VR image, generates accurate replacement demand in combination with user information;According to demand item index priority, efficient screening house matching information, then the evaluation value output result is calculated by containing user preference evaluation algorithm.Can solve the problem of traditional demand acquisition difficult, low matching granularity, improve matching efficiency and accuracy, guarantee housing matching reliability, also better meet the user residential replacement demand.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of residential information processing, and more particularly to a residential replacement matching system based on information analysis. BACKGROUND

[0002] With the development of individualization and diversification of residential needs, the residential replacement matching system based on information analysis has gradually become an important research direction in the field of housing management and optimization. However, the existing related technical solutions still have certain limitations in data processing, user demand analysis, matching accuracy and evaluation strategy, etc., and it is difficult to meet the efficient and accurate residential replacement needs.

[0003] A verification method, device and storage medium for matching houses and tenants are disclosed in CN113903091B. The patent sets parameters such as verification period, check-in times and check-in time period, and judges the matching result of tenants and houses by combining check-in information, thereby realizing accurate verification. However, this technical solution mainly focuses on the verification and matching of tenants and houses, lacks deep extraction and analysis of house feature information, and fails to fully consider the individualized needs of users and their dynamic changes in residential replacement. In addition, this solution does not provide a multi-dimensional evaluation mechanism for the matching result, which may result in insufficient accuracy of the matching result and user satisfaction. A model matching method, device and storage medium for house types are disclosed in CN118587461B. The patent divides the house type into multiple stations and uses point cloud data to generate an optimal rectangle for matching, solving the problem of manual marking in the matching of house type drawings and scanning data.

[0004] Since there is no effective means to obtain user needs, and filling in information through research and other methods cannot be quantified and accurately matched, how to accurately obtain user replacement needs has been a problem to be solved at present. The general method at present is to analyze user input information, user access behavior and user basic information to obtain user preferences. However, since the dimensions of house information are very many, it is difficult to match through language description or behavior analysis, and the existing house information has low granularity, making it difficult to accurately determine the matching of each house and user needs. SUMMARY

[0005] Therefore, the present application aims to provide a residential replacement matching system based on information analysis.

[0006] In order to solve the above technical problems, the technical solution of the present application is as follows: a residential replacement matching system based on information analysis, comprising a picture feature processing subsystem, a user demand analysis module, a replacement matching index module and a matching preference screening module.

[0007] The picture feature processing subsystem is configured with a feature extraction strategy, which is used to obtain a house VR image of a user to generate house feature information;

[0008] The user demand analysis module is configured with a demand generation strategy, which is used to generate replacement demand information according to the house feature information and the user information;

[0009] The replacement matching index module is configured with a demand index strategy, which sets a corresponding index priority for each demand item in the replacement demand information, generates a matching index clue according to the demand item corresponding to the index priority, and matches and obtains corresponding house matching information from the house information database according to the matching index clue, until the set of house matching information meets the matching condition, the house matching information includes the house feature information;

[0010] The matching preference screening module is configured with a matching evaluation strategy, which includes generating a matching demand parameter group and a user preference weight group according to the replacement demand information to construct a matching evaluation algorithm, and calculating a corresponding matching evaluation value of each house matching information through the matching evaluation algorithm, and outputting a matching result according to the matching evaluation value.

[0011] Firstly, through the VR image analysis method of the traditional platform, the user demand benchmark is obtained, based on the files uploaded by the user in the house platform, whether the user is not satisfied with the current house or replaces it for a certain purpose. Compared with information analysis and behavior analysis, VR image can have more dimensional possibilities in information refinement, so that user demand can be visualized and specific, and at the same time, through this way, the same dimensional analysis and quantification of housing resources can be realized, and through demand analysis and replacement matching, information can be quickly indexed, data reliability can be improved, and effective housing data can be matched efficiently. Based on the core technology of image feature recognition, it provides a reliable basis for matching users to adapt to housing resources under the condition of limited user information.

[0012] Further, the feature extraction strategy includes:

[0013] An image screening step, according to a preset picture extraction rule, extracts a picture data set from the VR image, the picture data set meets a preset feature abundance constraint, and the picture data set includes a plurality of house picture images;

[0014] An image labeling step, each house picture image is labeled by a type matching algorithm;

[0015] The feature extraction step is configured with a preset feature extraction sub-strategy, which is used to screen feature regions in the house picture image, and identify the feature information of each feature region through a feature recognition model to obtain feature sub-labels.

[0016] The VR video is split into a set of images through image screening, so that the volume, shape and other characteristics of the items can be extracted, and the picture image is labeled by type through a type matching algorithm, so that when the image features are identified, the corresponding position and region type are provided. Based on the type division, the identification efficiency is improved, and more associations can be established through the labeled relationship between the region and the image features. When demand analysis or matching is implemented, the identification dimension of the features is larger, and the demand between the features and the regions can be better reflected. Then, each furniture, item and decoration feature is identified by feature region, which improves the identification accuracy and ensures that all information in the image is collected and fed back as much as possible.

[0017] Further, the picture extraction rule includes acquiring continuous house picture images at a preset acquisition interval and an acquisition starting position, and adjacent house picture images satisfy a comprehensive constraint condition. The comprehensive constraint condition includes a coincidence constraint term, a feature continuity constraint term and a clarity constraint term. The coincidence constraint term reflects the coincidence between the picture images. The feature continuity constraint term reflects the continuity of the same feature region in the house picture. The clarity constraint condition reflects the clarity difference between adjacent pictures. The feature abundance constraint includes calculating the feature abundance value of each house picture image so that the average feature abundance value is greater than the feature abundance reference value.

[0018] By such a setting, the clarity, coincidence and feature continuity of the picture extraction rule are used to realize image processing of different VR videos, ensure that the information is fully reflected, and reduce the data volume to reduce the occupation of computing resources. Through the evaluation of the three dimensions, the extraction rule corresponding to each user is different according to the house characteristics and content, so that the user information can be fully extracted.

[0019] Further, the demand generation strategy includes:

[0020] A plurality of demand items are configured, and demand data corresponding to each demand item is matched from the house feature information according to the demand item;

[0021] Each demand item corresponds to an analysis sub-model, and the analysis sub-model is used to analyze the user information to obtain a demand adjustment value of the corresponding demand item;

[0022] The corresponding demand data is updated according to the demand adjustment value to generate the replacement demand information.

[0023] The demand generation strategy presents the demand characteristics of the aggregation user by configuring demand items, and different resolution sub-models are configured for different demand items, so as to split the demand of each dimension of the user, quantify the demand data of each dimension of the user, and present the demand data of each dimension of the user by vectorization. Different vector components are set to present the demand of the user in a certain dimension, and the resolution sub-model can adjust the original demand on the basis of the user's baseline demand combined with user information such as user input information, user basic information, user behavior information, etc.

[0024] Further, the demand items include storage demand items, living demand items, interference demand items, indoor demand items, environment demand items, cost demand items, and location demand items. The storage demand items reflect the storage demand of the user, the living demand items reflect the living area demand of the user, the interference demand items reflect the public area demand of the user, the indoor demand items reflect the indoor comfort demand of the user, the environment demand items reflect the living environment demand of the user, the cost demand items reflect the replacement cost demand of the user, and the location demand items reflect the living location demand of the user.

[0025] By configuring different configurations of the user's storage demand, living demand, interference demand, indoor demand, environment demand, cost demand, and location demand, the user can be analyzed based on the demand data. For example, the storage demand includes different types of storage, such as bathroom storage demand, washing and drying storage demand, and clothing storage demand, which can be quantified by setting different components. This can more accurately reflect the overall storage demand of the user. Similarly, other aspects of demand can also be set. The living demand can be the demand for several bedrooms and bedpans, and the interference demand can be the demand for public areas such as balcony space, toilet number, and dining space. The indoor demand includes factors such as decoration, lighting, ventilation, facility completeness, and courtyard area. The outdoor demand includes demands for corridor environment, hall environment, public area environment, and property. The cost demand includes demands for water and electricity fees, property fees, and rent. The location demand includes surrounding environment and public facilities. In this way, the user's demand can be fully and specifically refined. The basic information of the above demands is collected through a public platform based on VR images. Compared with traditional methods, it is difficult to collect information on a large number of demands, so this comparison work is generally completed together.

[0026] Further, the resolution sub-model is configured with an information grabbing library, the information grabbing library stores a plurality of information grabbing clues and corresponding feature triples, the resolution sub-model extracts the triple structure features corresponding to the feature triples from the user information according to the matching relationship of the information grabbing clues, and analyzes the triple structure features according to the user semantic preference library to obtain the corresponding demand adjustment value.

[0027] The information grabbing library configures information grabbing clues to configure each different information with different construction modes in the form of triples, for example, the triples corresponding to the wall surface are "material-texture-color", so that the characteristics of different types of clues can be refined by abstract extraction, and the purpose of quickly establishing different characteristic recognition modes can be met, and the pre-construction of the triple characteristic group ensures the information grabbing and recognition efficiency, meets the standardization of characteristics in information processing, and ensures the generation of demand adjustment values.

[0028] Further, the demand index strategy further includes a demand analysis sub-strategy, and the demand analysis sub-strategy is configured with a preference calculation algorithm, the preference calculation algorithm is used to calculate a user preference function of each demand item according to historical user behaviors in user information, and an index priority corresponding to the user preference function is generated according to the user preference function.

[0029] The demand index strategy includes a demand analysis sub-strategy, and the demand analysis sub-strategy analyzes the demand by configuring a preference calculation algorithm, analyzes a user preference function of each demand item, and generates an index corresponding priority value through the user preference function, so that the user's preference for different demand items can be identified and obtained through historical user behaviors, for example, the user generally browses decoration demand more than the scene where the user has lived, so it can be inferred that the user has a higher preference for decoration, and the user preference function can be corrected, and the user preference function can be indexed on the basis of a large amount of data to quickly identify and screen user demand.

[0030] Further, the matching condition is configured to have an abundance value according to the index priority of each demand item, and when the types of the house matching information corresponding to each demand item all meet the requirement of the abundance value, the matching condition is considered to be met, and the abundance value reflects the difference degree of all house matching information under the demand item.

[0031] Because the selected house information needs to have a certain diversity, the user can have different basis when selecting, on the one hand, the user demand can be further refined and updated through the user's selection behavior, and on the other hand, the house recognized by the algorithm can be as much as possible to be recognized by the user, saving the user's time to reselect.

[0032] Further, the matching evaluation strategy includes matching each demand item corresponding constituent sub-function from a preset function composition library according to user information, and generating a benchmark sub-function of the demand item according to the matching demand parameters in the matching demand parameter group, and the difference between the constituent sub-function and the benchmark sub-function is obtained to obtain a demand sub-function, and the demand sub-function corresponding to each user preference weight in the user preference weight group is weighted to construct the matching evaluation algorithm.

[0033] By matching the setting of the matching evaluation strategy, it is ensured that the user demand can be matched by the composed sub-function, for example, the demand of the user for different demand items is not linear, so according to the identification result, the preference of the user for a specific house is also not linear, so the composed sub-function can better reflect the specific preference of the user for a certain demand, for example, the cost brought by the rent, within a certain range, the user can accept, but if the range is exceeded, the demand may lead to other demands that will not enter the user selection even if they are matched, so through the calling of the composed sub-function, it is ensured that the user's information screening criteria are reasonable and reliable, and at the same time, the generated matching evaluation algorithm is consistent with the user's real demand screening logic, so as to accurately determine the actual needs of the user.

[0034] The technical effects of the present application mainly embody the following aspects: the system generates specific house characteristics with house VR images, generates accurate replacement demand in combination with user information, efficiently filters house matching information according to the index priority of demand items, and then calculates the evaluation value output result through the evaluation algorithm containing user preferences. It can solve the problems of traditional demand acquisition difficulty and low matching granularity, improve matching efficiency and accuracy, ensure the reliability of house source matching, and better meet the user's residential replacement demand. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The system overall architecture block diagram of the residential replacement matching system based on information analysis of the present application;

[0036] Figure 2 The feature extraction strategy flow chart of the residential replacement matching system based on information analysis of the present application;

[0037] Figure 3 The demand index strategy flow chart of the residential replacement matching system based on information analysis of the present application;

[0038] Figure 4 The matching evaluation strategy block diagram of the residential replacement matching system based on information analysis of the present application;

[0039] Figure 5 The matching condition judgment flow chart of the residential replacement matching system based on information analysis of the present application. DETAILED DESCRIPTION

[0040] The specific embodiments of the present application are further described in detail below in combination with the drawings, so that the technical scheme of the present application is easier to understand and master.

[0041] Referring to Figures 1-5 The residential replacement matching system based on information analysis, as shown in the figure, comprises a picture feature processing subsystem, a user demand analysis module, a replacement matching index module and a matching preference screening module.

[0042] With reference to Figure 2 As shown, the picture feature processing subsystem is configured with a feature extraction strategy for obtaining a user's house VR image to generate house feature information;

[0043] The feature extraction strategy includes:

[0044] An image screening step extracts a picture data set from the VR image according to a preset picture extraction rule, the picture data set satisfying a preset feature abundance constraint, and the picture data set including a plurality of house picture images; the picture extraction rule includes obtaining continuous house picture images at a preset acquisition interval and an acquisition starting position, adjacent house picture images satisfying a comprehensive constraint condition, the comprehensive constraint condition including a coincidence degree constraint term, a feature continuity constraint term, and a definition constraint term, the coincidence degree constraint term reflecting the coincidence degree between picture images, the feature continuity constraint term reflecting the continuity of the same feature map area in the house image picture, and the definition constraint condition reflecting the definition difference between adjacent pictures; and the feature abundance constraint includes calculating the feature abundance value of each house picture image to make the feature abundance mean value greater than the feature abundance reference value. The target of the image screening step is to extract a picture data set satisfying information integrity and effectiveness from the house VR image, and this step needs to strictly follow the preset picture extraction rule and feature abundance constraint, and the specific implementation is as follows: the picture extraction rule is realized by dynamically adjusting the acquisition interval and fixing the acquisition starting position. The acquisition starting position is uniformly set to 1 meter in front of the inside of the house entrance, and the extraction direction is clockwise through the whole house space; the acquisition interval is dynamically set according to the house space area, wherein the acquisition interval of a large space with an area ≥ 20 square meters is set to 0.5 seconds / frame, and the acquisition interval of a small space with an area < 20 square meters is set to 1 second / frame, so that the data redundancy can be reduced while ensuring that each space feature is not missed. The comprehensive constraint condition includes the coincidence degree constraint term, the feature continuity constraint term, and the definition constraint term, and the three conditions need to be met at the same time to retain the frame image, and the specific calculation method and constraint standard are as follows: the coincidence degree constraint term: the coincidence degree refers to the pixel ratio of the overlapping area in adjacent two house picture images, which is used to balance the information repetition and spatial continuity. The calculation formula is: , wherein represents the coincidence degree, represents the total number of pixels of the overlapping area in which the pixel values are completely consistent in adjacent two images, represents the total number of pixels of the previous image. The constraint standard of the coincidence degree in the system is , if the calculation result exceeds the range, the next frame image is removed and re-extracted until the constraint is met. The feature continuity constraint term: the feature continuity refers to the degree that the same house feature map area (such as the sofa in the living room, the bed in the bedroom, and the cabinet in the kitchen) is kept complete and position coherent in adjacent two images. The calculation formula is: wherein represents the feature continuity, refers to the number of complete pixels in the same feature region in adjacent two frames of images, refers to the total number of pixels of the feature region in a single frame of image. The constraint criterion of feature continuity in the system is , if the continuity of a feature region is lower than the value, it is determined that the frame of image cannot fully reflect the feature information and needs to be re-extracted. The definition of clarity is to avoid the fluctuation of clarity between adjacent two frames of images being too large to affect feature recognition. First, the Laplace operator method is used to calculate the clarity value of a single frame of image , and the specific method is to take the absolute value after Laplace convolution operation on the image, and then calculate the average value of all pixels as the G value of the frame; the calculation formula of the clarity difference value between adjacent two frames is: , wherein is the clarity value of the current frame, is the clarity value of the previous frame. The constraint criterion of clarity difference value in the system is ΔG≤10, if it exceeds the range, the playback rate of the VR video is adjusted and the image is re-extracted. The execution of feature abundance constraint: the feature abundance constraint is used to ensure that the picture data set can cover the core feature elements of each space of the house, and avoid the extracted image from causing incomplete feature recognition due to too little information. First, the "feature abundance value" is defined as the proportion of the number of actual recognized feature elements in a single frame of image to the total number of typical feature elements of the space, and the calculation formula is: , wherein E represents the feature abundance value of a single frame of image, refers to the number of actual recognized feature elements (such as sofa, tea table, bed, wardrobe, cabinet, window, etc.) in the frame of image, refers to the total number of preset typical feature elements of the space (a space feature database needs to be established in advance, for example, the of living room is set to 10, including sofa, tea table, TV, TV cabinet, lamp, wall decoration, window, curtain, floor, socket; the of bedroom is set to 8, including bed, wardrobe, bedside table, lamp, window, curtain, floor, socket; the of kitchen is set to 7, including cabinet, stove, sink, refrigerator site, lamp, floor, socket). The feature abundance average value is the arithmetic average value of the values of all extracted frames, and the calculation formula is: , wherein n is the total number of picture images extracted, to are the feature abundance values of the 1st frame to the frame, respectively. The constraint criterion of feature abundance average value in the system is ≥60%, if it is not satisfied, the number of image extraction frames of the corresponding space is increased until Up to standard, finally forming a qualified picture dataset.

[0045] Image labeling step, type matching algorithm is used to label the type of each house picture image; the goal of the image labeling step is to label the corresponding house space type for each frame of image in the picture dataset, so that the feature elements can be identified in combination with the space type during subsequent feature extraction. The specific implementation is as follows: selection and training of type matching algorithm: the system uses ResNet50 convolutional neural network model as the core of the type matching algorithm. This model is a commonly used image classification model in the field of computer vision, which has the characteristics of high recognition accuracy and high operation efficiency. The model training data uses the public House3D house space image dataset, which contains more than 100,000 house space images of different house types and different decoration styles, and has labeled “living room”, “master bedroom”, “secondary bedroom”, “kitchen”, “bathroom”, and “balcony” six types of space. In the training process, the dataset is divided into training set, validation set and test set according to the ratio of 7:2:1. The cross-entropy loss function is used as the optimization objective, and the stochastic gradient descent method (SGD) is used for model training. The number of training iterations is set to 50, the initial learning rate is set to 0.001 and decreases with the number of iterations, and finally the model's space type classification accuracy on the test set is ≥95%, ensuring the reliability of the labeling results.

[0046] Image labeling execution process: input the picture dataset obtained through the image screening step into the trained ResNet50 model, the model outputs the probability of the corresponding space type of each frame of image, and takes the space type with the highest probability as the labeling result of the frame of image; the labeling information is attached in the image file in the form of metadata, and the metadata format is “space type-extraction date-frame number”, for example, “living room-20240520-001” represents the first frame of living room image extracted on May 20, 2024, and “master bedroom-20240520-015” represents the 15th frame of master bedroom image extracted on May 20, 2024. After labeling, 10% of the images need to be manually inspected, if the accuracy rate of the inspection is consistent with the accuracy rate of the model test set, it is determined that the labeling is qualified; if there is a deviation, the model needs to be fine-tuned again and labeled again.

[0047] The feature extraction step is configured with a preset feature extraction sub-strategy for screening feature regions in the house picture image, and identifying feature information of each feature region through a feature recognition model to obtain a feature sub-label. The goal of the feature extraction step is to segment independent feature regions from the image of the labeled space type and identify the detailed information of each feature region to form a feature sub-label. The feature extraction sub-strategy is divided into two stages of "region segmentation" and "feature recognition". The first stage is region segmentation, which uses a U-Net semantic segmentation model to process the image of the labeled space type. The model needs to be trained in advance for different space type feature elements, for example, for a living room image, the model needs to be able to segment the areas corresponding to the sofa, coffee table, TV, etc. The segmentation accuracy requirement is that the pixel deviation of the feature region edge and the actual feature element edge is ≤5 pixels. If the deviation exceeds the range, adjust the segmentation threshold of the model and re-segment. Application of the feature recognition model: feature recognition uses a YOLOv8 target detection model to identify the information of each feature region obtained by region segmentation. The identification content includes three types of information: feature type, specification parameter, and attribute feature. Among them, the specification parameter needs to be calculated through the mapping relationship between image pixels and actual size. The mapping formula is: wherein represents the actual size of the feature element, represents the pixel length of the feature region in the image, represents the actual distance between the camera and the feature element when the VR video is taken, represents the focal length of the camera. The attribute feature includes color, material, etc., which is obtained by analyzing the texture and color features of the feature region through the model, for example, "gray fabric sofa" and "light brown wooden coffee table".

[0048] Generation of feature sub-labels: The identification results of each feature region are summarized as feature sub-labels. The format of the feature sub-label is "space type-feature type-specification parameter-attribute feature", for example, "living room-fabric sofa-length 2.2 meters-gray" "kitchen-wooden cabinet-height 2.4 meters-white" "bedroom-wooden bed-width 1.8 meters-dark brown". All feature sub-labels are stored by space type, and finally form a complete house feature information, which contains all the details of the feature elements of each space of the house, and can be directly used for subsequent user demand analysis.

[0049] The core function of the picture feature processing subsystem is to convert the house VR image uploaded by the user into structured and quantifiable house feature information, to generate replacement demand information for the subsequent user demand analysis module, and to provide accurate and comprehensive data support for the house matching information screening of the replacement matching index module. The subsystem completes the image processing, space type marking and feature element recognition of the VR image in steps through the preset feature extraction strategy. The entire process needs to meet the operability and repeatability to ensure that the system deployment and operation can be realized by the technical personnel in the technical field according to the following content.

[0050] Referring to Figure 3 The user demand analysis module is configured with a demand generation strategy for generating replacement demand information according to house feature information and user information. The core function of the user demand analysis module is to receive the house feature information output by the picture feature processing subsystem and generate structured and quantifiable replacement demand information in combination with user information. Meanwhile, the index priority of each demand item is determined through a demand analysis sub-strategy to provide accurate demand basis for the subsequent replacement matching index module.

[0051] The demand generation strategy includes:

[0052] The demand generation strategy includes:

[0053] Each demand item corresponds to an analysis sub-model for analyzing user information to obtain a demand adjustment value of the corresponding demand item.

[0054] The analysis sub-model is configured with an information extraction library storing a plurality of information extraction clues and corresponding feature triples. The analysis sub-model extracts the triple structure feature corresponding to the feature triples from the user information according to the matching relationship of the information extraction clues, and analyzes the triple structure feature according to the user semantic preference library to obtain the corresponding demand adjustment value. The core of the analysis sub-model is to extract the key information affecting the demand from the user information, calculate the demand adjustment value to correct the baseline demand data, and the execution steps are as follows:

[0055] The construction process of the information extraction library is as follows: the information extraction library needs to pre-store information extraction clues and feature triples corresponding to the seven demand items and sub-dimensions. The information extraction clue is a keyword or behavior identifier for identifying demand-related information from user information, for example, the extraction clues of the storage demand item include "bedroom", "cabinet", "handle", "open", and other identification features, so as to facilitate the system to lock the type for feature extraction during image processing; the feature triple is a three-group of associated information used to structurally describe the key attributes of the demand item, in the format of "demand dimension-property type-current state", for example, the feature triple of the storage demand item can be set as "clothing storage-space capacity-1.2 cubic meters"; in addition, if the wardrobe is in a closed state, "clothing storage-projection capacity-1 square meter".

[0056] Extraction and analysis of feature triples: according to the matched information extraction clues, the corresponding feature triples are called from the information extraction library, and the three-dimensional structure features are analyzed in combination with the user semantic preference library to determine the demand expectation. The user semantic preference library is a pre-constructed database containing semantic mapping relationships, for example, "increasing storage space" is mapped to "demand expectation capacity is 30% higher than the current one", and "near subway" is mapped to "commuting time ≤ 20 minutes".

[0057] Calculation of demand adjustment value: the demand adjustment value adopts a linear calculation model, and the formula is wherein represents the demand adjustment value of a demand item sub-dimension, represents an adjustment coefficient, represents the demand expectation obtained from the user semantic preference library, represents the benchmark demand data matched from the house feature information. The adjustment coefficient is set according to the user's attention degree to the demand item, and the attention degree is determined by the user's mention times or behavior frequency, for example, if the user mentions "increasing clothing storage" for 3 times, then 1.2 is taken, and only once is mentioned 1.0 is taken, and if there is no mention but there is relevant browsing behavior, then a is taken as 0.8, The value range of a is 0.8-1.2 to ensure that the adjustment range is reasonable.

[0058] The demand items include storage demand items, living demand items, interference demand items, indoor demand items, environment demand items, cost demand items, and location demand items, the storage demand items reflect the storage demand of the user, the living demand items reflect the living area demand of the user, the interference demand items reflect the public area demand of the user, the indoor demand items reflect the indoor comfort demand of the user, the environment demand items reflect the living environment demand of the user, the cost demand items reflect the replacement cost demand of the user, and the location demand items reflect the living location demand of the user.

[0059] The demand generation strategy first needs to split the preset seven demand items into sub-dimensions. Each demand item sub-dimension needs to correspond to the feature sub-label in the housing feature information and the associated dimension of the user information, ensuring that the demand data can be accurately matched. The specific details are as follows:

[0060] Storage demand items: including bathroom storage sub-dimension, laundry storage sub-dimension, and clothing storage sub-dimension. The bathroom storage sub-dimension corresponds to feature sub-labels such as "bathroom - storage rack - capacity 0.3 cubic meters" and "bathroom - mirror cabinet - capacity 0.2 cubic meters" in the housing feature information. The laundry storage sub-dimension corresponds to feature sub-labels such as "balcony - laundry cabinet - capacity 0.5 cubic meters" and "balcony - clothesline - load capacity 5 kg". The clothing storage sub-dimension corresponds to feature sub-labels such as "bedroom - wardrobe - volume 1.2 cubic meters" and "bedroom - five-drawer cabinet - volume 0.4 cubic meters".

[0061] Residential demand items: including bedroom number sub-dimension, bedroom bed size sub-dimension, and residential population adaptation sub-dimension. The bedroom number sub-dimension corresponds to the number of space type labels such as "master bedroom" and "secondary bedroom" in the housing feature information. The bedroom bed size sub-dimension corresponds to feature sub-labels such as "bedroom - solid wood bed - width 1.8 meters". The residential population adaptation sub-dimension is determined in combination with the family size in the user information, such as a 3-person family requiring at least 2 bedrooms.

[0062] Interference demand items: including public area area sub-dimension, bathroom number sub-dimension, and dining space sub-dimension. The public area area sub-dimension is calculated by the area of "living room" and "balcony" in the housing feature information. The bathroom number sub-dimension corresponds to the number of space type labels such as "bathroom". The dining space sub-dimension corresponds to feature sub-labels such as "living room - dining table - floor area 0.8 square meters".

[0063] Indoor demand items: including lighting sub-dimension, ventilation sub-dimension, decoration style sub-dimension, and facility completeness sub-dimension. The lighting sub-dimension corresponds to feature sub-labels such as "living room - window - area 1.5 square meters" and "bedroom - window - orientation south". The ventilation sub-dimension corresponds to the number of windows and the opening method, such as "living room - window - sliding type". The decoration style sub-dimension corresponds to attribute features such as "living room - wall - latex paint - white" and "floor - solid wood - light brown". The facility completeness sub-dimension corresponds to feature sub-labels such as "kitchen - gas stove - brand XX" and "bathroom - water heater - type electric water heater".

[0064] Environmental demand items: including corridor environment sub-dimension, public area environment sub-dimension, and property service sub-dimension. The corridor environment sub-dimension is extracted from user information such as "mentioning corridor cleanliness". The public area environment sub-dimension corresponds to expressions such as "hope the community has greenery" in the user information. The property service sub-dimension corresponds to the behavior of clicking on "24-hour security" in the user's historical browsing records.

[0065] Cost requirements include sub-dimensions for utilities and property management fees, rent, and replacement fees. The utilities and property management fees sub-dimension is extracted from the user's input text "monthly utilities and property management fees not exceeding 500 yuan." The rent sub-dimension corresponds to the rent range of properties the user has historically viewed (e.g., 1500-2000 yuan / month). The replacement fees sub-dimension corresponds to the user's statement "desiring the fees to be less than 1% of the total house price." Alternatively, the user's cost requirements can be extracted from the current property information.

[0066] Location requirements include the following sub-dimensions: commuting distance, surrounding amenities, and regional preference. The commuting distance sub-dimension is extracted from the user information, which states that "commuting to A1 company takes no more than 30 minutes." The surrounding amenities sub-dimension corresponds to text such as "hoping for supermarkets and schools nearby." The regional preference sub-dimension corresponds to the user's historical browsing history of properties in the "A3 area."

[0067] It should be noted that there can be multiple ternary feature groups for each item or feature. For example, they can represent color, material, or position, without limitation.

[0068] The baseline demand data for each sub-dimension of demand items are added to the demand adjustment value to obtain the corrected demand target value. All demand target values ​​are summarized by demand item to form the final replacement demand information.

[0069] The demand parsing sub-strategy calculates the user's preference for each demand item using a preference calculation algorithm, generating index priorities to guide subsequent replacement matching. The core of the preference calculation algorithm is constructing a user preference function, the formula of which is... ,in Indicates the user's opinion on the first The preference value of each demand item. Indicates the weight of the number of interactions. Indicates the weight of dwell time. This represents the number of times a user interacts with information related to the i-th request item. Indicates that the user browsed the first The total dwell time on information related to each demand item This represents the normalized coefficient representing the number of interactions and dwell time for all request items. , This represents the total number of interactions for all requirement items. This indicates the total dwell time for all demand items. and The value is always 0.5 to ensure that the number of interactions and the duration of the interaction are equally important, preference value The value ranges from 0 to 1; a larger value indicates a higher degree of user preference for that requirement. The index priorities of each demand item are sorted in descending order.

[0070] Referring to Figure 3 As shown, the permutation matching index module is configured with a demand index strategy, the demand index strategy sets a corresponding index priority for each demand item in the permutation demand information, and generates a matching index clue according to the demand item corresponding to the index priority, and matches the corresponding housing matching information from the housing information database according to the matching index clue, until the set of housing matching information meets the matching condition, the housing matching information includes the housing feature information; the demand index strategy further includes a demand analysis sub-strategy, the demand analysis sub-strategy is configured with a preference calculation algorithm, the preference calculation algorithm is used to calculate a user preference function of each demand item according to historical user behavior in the user information, and generate a corresponding index priority according to the user preference function.

[0071] The core function of the permutation matching index module is to undertake the permutation demand information output by the user demand analysis module and the index priority of each demand item, generate a matching index clue through a preset demand index strategy, retrieve and filter the housing matching information from the housing information database, until the set of housing matching information corresponding to all demand items meets the matching condition, provide candidate housing data that meets the demand for the subsequent matching preference filtering module, wherein the housing matching information contains the housing feature information generated by the picture feature processing subsystem.

[0072] Referring to Figure 5As shown, matching index clues refer to the set of filtering conditions generated based on the target value and index priority of each requirement item, used to accurately retrieve housing listings in the housing information database. They need to correspond one-to-one with the fields in the housing information database (such as commuting time, storage space, property fees, etc.). The specific steps of the demand indexing strategy are as follows: First, determine the priority of demand items. The preference values ​​output by the user demand parsing module are used for sorting. Demand items with higher preference values ​​have higher index priority. During retrieval, priority is given to filtering based on high-priority demand items to reduce invalid search data. Second, generate matching index clues. Filtering conditions are constructed for each demand item's sub-dimensional target value, ranked from highest to lowest priority. For example, if the highest priority location demand item's target value is "commuting time ≤ 22 minutes and there is a primary school within 3 kilometers," then the corresponding matching index clue is "commuting time field ≤ 22 minutes AND surrounding facilities field includes a primary school." Clues for lower-priority demand items need further filtering based on the search results of higher-priority clues to avoid redundant cross-priority searches. Third, execute database retrieval. The generated matching index clues are sequentially input into the housing information database. Each round of retrieval only retains housing data that matches the current clue, serving as the base data pool for the next round of low-priority clue retrieval. Fourth, determine if the matching conditions are met. Matching conditions need to be verified by calculating the abundance value of each demand item. The abundance value is an indicator reflecting the difference in matching information among all housing items under a certain demand item. The calculation formula is... ,in Indicates the first Abundance values ​​of each demand item This indicates the number of housing matching information types with differentiated characteristics under this demand item. This indicates the total number of housing matching information corresponding to this demand item in the current data pool. This system sets the abundance value of all demand items to be ≥30% to be considered as meeting the matching conditions. If the abundance value of a certain demand item does not meet the standard, it is necessary to supplement the search for other housing listings corresponding to the demand item clues until the abundance value meets the requirements. At this time, the search stops and all housing matching information in the current data pool is output.

[0073] The matching conditions are configured with an abundance value based on the index priority of each demand item. When the types of housing matching information corresponding to each demand item meet the abundance value requirements, the matching conditions are considered to be met. The abundance value reflects the degree of difference among all housing matching information under the demand item.

[0074] The matching preference screening module is configured with a matching evaluation strategy, the matching evaluation strategy includes generating a matching demand parameter group and a user preference weight group according to the replacement demand information to construct a matching evaluation algorithm, and calculating a matching evaluation value corresponding to each housing matching information through the matching evaluation algorithm, and outputting a matching result according to the matching evaluation value. The matching evaluation strategy includes matching each demand item corresponding component sub-function from the preset function component library according to the user information, and generating a benchmark sub-function of the demand item according to the matching demand parameter in the matching demand parameter group, and obtaining a demand sub-function by subtracting the component sub-function and the benchmark sub-function, and constructing the matching evaluation algorithm by weighting the corresponding demand sub-function according to each user preference weight in the user preference weight group.

[0075] The core function of the matching preference screening module is to accept the housing matching information output by the replacement matching index module, combine the replacement demand information generated by the user demand analysis module, construct a matching evaluation algorithm through a preset matching evaluation strategy, calculate the matching evaluation value of each set of housing matching information, and finally output the matching result according to the evaluation value., provide candidate housing sources that meet the individual preferences for users, the whole process needs to ensure that the algorithm logic is reproducible and the calculation steps are operable to meet the implementation needs of the technical personnel in the technical field.

[0076] The matching demand parameter group refers to the target value set of each demand item sub-dimension extracted from the replacement demand information, and each demand item corresponds to at least one parameter. The user preference weight group refers to the weight set allocated in proportion according to the preference value of each demand item output by the user demand analysis module, and the total weight is 1, which is used to reflect the importance of different demand items in user decision-making, for example, the user's preference value for the location demand item is the highest, and its weight can be set to 0.3, the storage demand item is second, and the environment and cost demand items are set to 0.2 and 0.25 respectively. The function component library refers to a database that stores the component sub-functions corresponding to each demand item, and the component sub-function is a mathematical function that reflects the actual preference of the user for a certain demand item, which needs to be matched according to the demand item type, for example, the cost demand item corresponds to a segmented function, and the environment demand item corresponds to a classification assignment function. Technical personnel can pre-configure functions according to common user preference types, or can iterate and optimize function forms through user historical behavior data.

[0077] First, extract the matching demand parameter group: screen the sub-dimension target value of each demand item from the replacement demand information one by one, eliminate duplicate or redundant parameters, and form a structured parameter group. For example, user B obtains demand items respectively 、 、 、 、 、 、 , corresponding to storage, residence, interference, indoor, environment, cost, and location.

[0078] Second step, generate user preference weight group: based on the user demand analysis module output of each demand item preference value, using the normalization calculation weight, the formula is , wherein represents the weight of the th demand item, represents the preference value of the th demand item, represents the sum of all demand item preference values. For example, the location demand item preference value of user B is , , , , , , .

[0079] Third step, according to the user information to obtain each component sub function, the component sub function is obtained through the user behavior, and then the corresponding reference sub function is generated according to the user demand vector, and the matching evaluation algorithm is constructed:

[0080] + + + + + + , wherein is the demand item corresponding to the storage sub function, is the storage demand vector deviation value, reflecting the actual storage demand and the demand deviation of the corresponding house source, is the demand item corresponding to the living sub function, is the living demand vector deviation value, reflecting the actual living demand and the demand deviation of the corresponding house source, is the demand item corresponding to the interference sub function, is the interference vector deviation value, reflecting the actual interference and the demand deviation of the corresponding house source, is the demand item corresponding to the indoor sub function, is the indoor demand vector deviation value, reflecting the actual indoor demand and the demand deviation of the corresponding house source, is the demand item corresponding to the environment sub function, is the environment demand vector deviation value, reflecting the actual environment demand and the demand deviation of the corresponding house source, is the demand item corresponding to the cost sub function, is a cost demand vector deviation value, reflecting the actual cost demand and the demand deviation of the corresponding house source, is a demand item is a corresponding position sub-function, is a position demand vector deviation value, reflecting the actual position demand and the demand deviation of the corresponding house source; by setting in this way, the information of each house source is quantified, and then the deviation can be brought in to calculate the matching result, and the result value corresponding to each house source is sorted to realize the information comparison output.

[0081] Of course, the above is only a typical example of the present application, in addition to this, the present application can have other various specific embodiments, and any technical solution formed by equivalent replacement or equivalent transformation falls within the scope of the present application.

Claims

1. An information analysis based residential replacement matching system, characterized by: The system comprises a picture feature processing subsystem, a user demand analysis module, a replacement matching index module, and a matching preference screening module. The picture feature processing subsystem is configured with a feature extraction strategy for obtaining a user's house VR image to generate house feature information. The user demand analysis module is configured with a demand generation strategy for generating replacement demand information according to the house feature information and user information. The replacement matching index module is configured with a demand index strategy for setting an index priority for each demand item in the replacement demand information, generating a matching index clue according to the index priority corresponding demand item, and matching corresponding house matching information from a house information database according to the matching index clue until the set of house matching information meets the matching condition, wherein the house matching information comprises the house feature information. The matching preference screening module is configured with a matching evaluation strategy comprising generating a matching demand parameter group and a user preference weight group according to the replacement demand information to construct a matching evaluation algorithm, calculating a corresponding matching evaluation value of each house matching information through the matching evaluation algorithm, and outputting a matching result according to the matching evaluation value.

2. The information analysis based residential relocation matching system of claim 1, wherein: The feature extraction strategy comprises: An image screening step for extracting a picture data set from the VR image according to a preset picture extraction rule, wherein the picture data set meets a preset feature abundance constraint, and the picture data set comprises a plurality of house picture images; An image labeling step for labeling each house picture image by type through a type matching algorithm; A feature extraction step configured with a preset feature extraction sub-strategy for screening feature image areas in the house picture images, and identifying feature information of each feature image area through a feature recognition model to obtain feature sub-labels.

3. The information analysis based residential relocation matching system of claim 2, wherein: The picture extraction rule comprises obtaining continuous house picture images at a preset acquisition interval and acquisition starting position, wherein adjacent house picture images meet a comprehensive constraint condition, the comprehensive constraint condition comprises a coincidence degree constraint item, a feature continuity constraint item, and a definition constraint item, the coincidence degree constraint item reflects the coincidence degree between picture images, the feature continuity constraint item reflects the continuity of the same feature image area in the house image picture, and the definition constraint condition reflects the definition difference between adjacent pictures; the feature abundance constraint comprises calculating a feature abundance value of each house picture image to make the feature abundance mean value greater than a feature abundance benchmark value.

4. The information analysis based residential relocation matching system of claim 2, wherein: The demand generation strategy comprises: Configured with a plurality of demand items, each demand item is matched with corresponding demand data from the house feature information; Each demand item corresponds to an analysis sub-model for analyzing user information to obtain a demand adjustment value of the corresponding demand item; The corresponding demand data is updated according to the demand adjustment value to generate the replacement demand information.

5. The information analysis based residential relocation matching system of claim 4, wherein: The demand items include a storage demand item, a living demand item, an interference demand item, an indoor demand item, an environment demand item, a cost demand item, and a location demand item, the storage demand item reflecting a storage demand of the user, the living demand item reflecting a living area demand of the user, the interference demand item reflecting a public area demand of the user, the indoor demand item reflecting an indoor comfort demand of the user, the environment demand item reflecting a living environment demand of the user, the cost demand item reflecting a replacement cost demand of the user, and the location demand item reflecting a living location demand of the user.

6. The information analysis based residential relocation matching system of claim 4, wherein: The analysis sub-model is configured with an information extraction library, the information extraction library storing a plurality of information extraction clues and corresponding feature triplets, the analysis sub-model extracting a triplet structure feature corresponding to the feature triplets from the user information according to a matching relationship of the information extraction clues, and analyzing the triplet structure feature according to the user semantic preference library to obtain a corresponding demand adjustment value.

7. A residential relocations matching system based on information analysis as claimed in claim 6 wherein: The demand index strategy further includes a demand analysis sub-strategy, the demand analysis sub-strategy being configured with a preference calculation algorithm, the preference calculation algorithm being used to calculate a user preference function of each demand item according to historical user behaviors in the user information, and generating a corresponding index priority according to the user preference function.

8. The information analysis based residential relocation matching system of claim 1, wherein: The matching condition is configured to have an abundance value according to the index priority of each demand item, when types of the house matching information corresponding to each demand item all meet the requirement of the abundance value, the matching condition is considered to be met, and the abundance value reflects a difference degree of all house matching information under the demand item.

9. The information analysis based residential relocation matching system of claim 1, wherein: The matching evaluation strategy includes matching a component sub-function corresponding to each demand item from a preset function component library according to the user information, generating a benchmark sub-function of the demand item according to a matching demand parameter in a matching demand parameter group, and obtaining a demand sub-function by subtracting the component sub-function from the benchmark sub-function, weighting the demand sub-function corresponding to each user preference weight in a user preference weight group according to the user preference weight to construct the matching evaluation algorithm.

Images