Information recommendation method and device, storage medium and program product

Abstract

The embodiment of the invention provides an information recommendation method and device, a storage medium and a program product. In the embodiment of the invention, the target geographic object associated with the user and the attribute data of the target geographic object in the target library are introduced, and the geographic semantic feature, the distance feature and the geographic context feature are incorporated into the user modeling process, so that the data normal form which only depends on established behavior data and static portraits is broken through; on this basis, according to attribute data and historical behavior data of the target geographic object, steady-state demand potential information and dynamic interest intention information of the user are predicted, and user demand features are described from two dimensions of long-term steady-state demand and short-term dynamic interest; furthermore, the steady-state demand potential information and the dynamic interest intention information are combined with candidate information objects in the target application scene to perform joint information recommendation, so that the recommendation result can not only fit the current interest of the user, but also keep the consistency with the long-term demand of the user, and the recommendation effect is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to an information recommendation method and device, a storage medium and a program product. BACKGROUND

[0002] In the field of e-commerce, the personalized recommendation system based on user portrait and historical behavior data has become the core infrastructure for improving platform operation efficiency and user experience. The current mainstream recommendation technology, whether collaborative filtering, content analysis, or more complex deep learning hybrid models, its recommendation effect largely depends on the input data dimensions, such as dynamic behavior sequence, static basic portrait, etc.

[0003] Among them, the dynamic behavior sequence: such as user click, browse, search, purchase record and other behavior data, constitutes the most direct expression of user intention, and is the core source of capturing user intention for the recommendation system. The static basic portrait: such as the attribute information in the user registration information or the interest label inferred through behavior, can provide relatively stable prior knowledge for the recommendation system.

[0004] However, these data dimensions, while constituting the basis of the recommendation system's ability, also solidify its cognitive boundaries. The deep dependence on this limited data set makes the recommendation effect trapped in an "ceiling" that is difficult to break through, forming a technical barrier for its own development.

[0005] However, this deep dependence on the given data dimensions also leads to the "involution" of the recommendation system. Within the existing data paradigm, the optimization space of the recommendation system is increasingly narrowing, making it difficult to achieve a step-by-step improvement in recommendation effect. SUMMARY

[0006] The embodiments of the present application provide an information recommendation method, device, storage medium and program product, to break through the traditional data paradigm for recommendation and greatly improve the recommendation effect of the recommendation system.

[0007] The embodiment of the present application provides an information recommendation method, comprising: obtaining portrait data and historical behavior data of a user, wherein the portrait data comprises identification information of a target geographical object associated with the user; obtaining attribute data of the target geographical object from a target library according to the identification information, wherein the target library is obtained by performing data enhancement on geographical object description information in an ontology library based on geographical semantic features, distance features and geographical context features of the geographical object description information, and the geographical object description information comprises attribute data of a corresponding geographical object; predicting steady-state demand potential information of the user according to the attribute data of the target geographical object, and predicting dynamic interest intention information of the user according to the historical behavior data; and performing joint information recommendation on the user according to the steady-state demand potential information and the dynamic interest intention information of the user in combination with candidate information objects in a target application scenario.

[0008] The embodiment of the present application also provides an electronic device, comprising a memory and a processor, wherein the memory is used for storing a computer program, and the processor is coupled with the memory and used for executing the computer program to implement steps in the above method.

[0009] The embodiment of the present application also provides a computer readable storage medium storing a computer program, which enables the processor to implement steps in the above method when the computer program is executed by the processor.

[0010] The embodiment of the present application also provides a computer program product, which comprises computer programs / instructions, and enables the processor to implement steps in the above method embodiment when the computer programs / instructions are executed by the processor.

[0011] In the embodiment of the present application, the target geographical object associated with the user and attribute data thereof in the target library are introduced, geographical semantic features, distance features and geographical context features are included in the user modeling process, and the data paradigm of only relying on given behavior data and static portrait is broken through; on this basis, steady-state demand potential information and dynamic interest intention information of the user are predicted according to the attribute data of the target geographical object and the historical behavior data respectively, and the demand characteristics of the user are described from two dimensions of long-term steady-state demand and short-term dynamic interest; further, the steady-state demand potential information and the dynamic interest intention information are combined with candidate information objects in a target application scenario to perform joint information recommendation, so that the recommended result can not only fit the current interest of the user, but also maintain consistency with the long-term demand of the user, and the recommendation effect is greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0012] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings: Figure 1 is a first application scenario diagram of a geographic object value evaluation method provided by a first embodiment of the application; Figure 2 is a second application scenario diagram of a geographic object value evaluation method provided by the first embodiment of the application; Figure 3 is a flowchart of a geographic object value evaluation method provided by the first embodiment of the application; Figure 4 is a curve fitting diagram in a geographic object value evaluation method provided by the first embodiment of the application; Figure 5 is a schematic diagram of a geographic object value evaluation device provided by a second embodiment of the application; Figure 6 is a schematic diagram of an electronic device provided by a third embodiment of the application; Figure 7 is a flowchart of an information recommendation method provided by an embodiment of the application; FIG. 8 is a flowchart of another information recommendation method provided by an embodiment of the application; Figure 9 is a structural schematic diagram of an electronic device provided by an embodiment of the application. DETAILED DESCRIPTION

[0013] To make the objectives, technical solutions, and advantages of the application clearer, the following will describe the technical solutions of the application in conjunction with specific embodiments of the application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the application but not all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the application.

[0014] It should be noted that in the case where the embodiments of the application involve user information, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the embodiments of the application are all information and data authorized by the user or authorized by all parties, and the collection, use, and processing of related data need to comply with relevant laws, regulations, and standards of relevant countries and regions, and provide corresponding operation portals for users to choose authorization or refusal. In addition, the various models (including but not limited to language models or large models) involved in the application comply with relevant legal and standard regulations.

[0015] The valuation of a geographic object (such as a property) refers to the evaluation of resource data required for a title transfer event of the geographic object, and is a key technology for realizing asset digitization. The accuracy of the valuation of the geographic object is of great significance to the transaction of the geographic object, and is an important prerequisite for the digitization of asset transactions. Currently, when evaluating the value of a geographic object, one is to rely on manual calculation to evaluate the value of the geographic object. Manual valuation not only has a long cycle and high cost, but also has a large evaluation difference due to the experience and preference of the valuer. The other is to use a geographic object valuation model to estimate the value of the geographic object. Although the use of digital form to estimate the value of the geographic object can improve the accuracy of the evaluation, when using the existing geographic object valuation model to evaluate the value of the geographic object, a large amount of disordered related data needs to be collected, and the collected data will be different due to different users. This leads to the fact that this method not only occupies a large amount of computing resources, is time-consuming and laborious, and moreover, there is a large difference between the valuation result of the geographic object and the actual transaction result, and between the use results of different users, which leads to the situation that the transaction cycle of the geographic object is too long or cannot be completed. Therefore, how to improve the efficiency of the valuation of the value of the geographic object while reducing the computing resources is a problem that needs to be solved at present.

[0016] Therefore, the embodiment of the present application provides a geographic object value evaluation method. When obtaining the value evaluation result of a target geographic object, the method comprehensively considers the preliminary value evaluation result determined based on the attribute data of the target geographic object, the statistical value of the historical transaction prices of all similar geographic objects similar to the target geographic object, the prediction value of the evaluation result of the current region determined by the fitting curve of the historical transaction prices and time relationship of other geographic objects in the region of the target geographic object at different times, the trend change coefficient determined by the ratio of the prediction value to the historical transaction price corresponding to the previous time unit on the fitting curve, and the influence of the missing degree of the similar geographic object on the value evaluation result of the target geographic object. The accuracy of the value evaluation of the target geographic object is improved, and the data required for the evaluation calculation is ordered and determined, which reduces the consumption of computing resources, improves the efficiency of the evaluation calculation, and improves the stability of the calculation result.

[0017] The following explains the technical terms related to the embodiments of the present application: Robustness: refers to the stability and adaptability of a system, model or method in the face of uncertainty, interference or changes. A robust system can maintain its performance under different conditions without being affected by external factors. Robustness can be applied in many fields, including engineering, statistics, computer science, artificial intelligence, etc.

[0018] Unsupervised learning: A paradigm of machine learning that primarily differs from supervised learning in that it does not use labeled data. In unsupervised learning, algorithms learn from unlabeled data, trying to discover patterns or structures within the data. Common tasks in unsupervised learning include clustering, dimensionality reduction, and anomaly detection, among others.

[0019] Time series: A dataset that records a series of observations in chronological order. These observations can be any measurable variable, such as stock prices, temperature changes, website visits, and more. Time series analysis has wide applications in various fields, including but not limited to economics, finance, meteorology, engineering, and social sciences.

[0020] The following Figures 1-2 The application scenarios of the geographic object value evaluation method provided by the embodiments of the present application are described in detail.

[0021] Please refer to Figure 1 , Figure 1 is a first application scenario diagram of a geographic object value evaluation method provided by the first embodiment of the present application. In the embodiments of the present application, a target geographic object is taken as a house source for example to describe the evaluation of the value of the house source. First, attribute data of the house source is obtained, which can be information of a community where the house source is located, such as the number of buildings in the community, the construction year of the community, and the greening information of the community; then, house source information is obtained, such as the floor where the house source is located, the area of the house source, and the historical transaction prices of other house sources in the community; finally, geographic information of the house source is also obtained, such as the zoning information to which the house source belongs, whether there are subways, parks, schools, and other supporting facilities around the house source, and the like. The attribute data of the house source is comprehensively obtained, and a preliminary value evaluation result of the house source, i.e., a first result, is obtained through a preset regression model. Then, clustering information is obtained through a clustering model to obtain similar house sources similar to the attribute data of the house source to be evaluated, further obtain historical transaction prices of all similar house sources, and calculate statistical values of all historical transaction prices as a second result of similar prediction. And according to the number of similar house sources and the number of transaction house sources in the region where the house source to be evaluated is located, the missing degree of the similar house sources is determined. Then, based on the prices of the community where the house source is located, the prices of the business circle, the prices of the zoning, and the prices of the city, a trend change coefficient of the house source is obtained based on a time series model. Finally, the missing degree is used as an adjustment coefficient, the product of the second result and the trend change coefficient is used as an adjustment amount, the first result is adjusted, and finally the value evaluation result of the house source is obtained.

[0022] In specific implementation, the first additional result of the house to be evaluated can also be obtained according to the category data and the attribute data of the house to be evaluated, the second additional result of the house to be evaluated can also be obtained according to the city data and the attribute data of the house to be evaluated, and finally, the total evaluation result of the house to be evaluated can be obtained by fusing the value evaluation result, the first additional result and the second additional result. Please refer to Figure 2 , Figure 2 is a second application scenario diagram of a geographic object value evaluation method provided by the first embodiment of the present application. In the embodiment of the present application, the geographic objects can be classified, for example, the types of the geographic objects can be houses, apartments, villas, or garages. The cities where the geographic objects belong to can also be classified, for example, the geographic objects are in a first-tier city or a second-tier city, in the east of China or in the northwest of China, etc. It should be noted that each type corresponds to a sub-model, by inputting the attribute data of the above different types into the corresponding sub-model, the additional evaluation result of the target geographic object can be obtained, and the total evaluation result of the house can be obtained according to the value evaluation result and the additional evaluation result. Of course, in specific implementation, the total evaluation result of the house can also be corrected, for example, in the scenario of sudden market fluctuation, the influence degree of market fluctuation on the value evaluation result is determined as a correction coefficient, and the above value evaluation result or total evaluation result can be corrected by using the correction coefficient, so that the final estimated result of the house is more accurate.

[0023] The geographic object value evaluation method described in the embodiments of the present application mainly includes three parts, the attribute layer, the model layer and the strategy layer. The attribute layer is the foundation, which can subdivide the housing according to multiple dimensions such as residential attributes and geographic locations, so as to select a suitable evaluation model. Then, different types of housing are routed to the most suitable main and sub-models, thereby improving the accuracy and efficiency of evaluation. Among them, the residential attributes include building age, area, house type, floor, etc., which can reflect the basic situation of the housing and are important factors for evaluating the value of the housing. The geographic location mainly considers the neighborhood, district, and even city where the housing is located, which has a significant impact on housing prices, such as areas with convenient transportation and commercial prosperous districts. The model layer is the core part, which is composed of an evaluation main model and multiple sub-models. The evaluation main model as the core model integrates multiple key factors such as historical transactions, similar housing, market trends, etc., and its purpose is to provide a preliminary and comprehensive evaluation reference. The sub-model carefully considers the unique attributes of specific types of housing. In a complex real estate market, a single model is difficult to adapt to all situations. Therefore, multiple sub-models are customized according to different residential attributes and geographic locations. For example, the apartment sub-model may specially consider the special attributes of the apartment, while the city sub-model pays more attention to the city level to which the housing belongs. Although each sub-model may be similar in structure, it will be significantly different in the processing of specific factors to ensure that its prediction for specific housing types is more accurate. The strategy layer is mainly to finally process the results of the model, which is responsible for further correction processing of the evaluation results output by the model layer, to ensure the robustness and flexibility of the system. The evaluation results output by the model may be affected by various non-normal factors in actual application, such as sudden market fluctuations. The strategy layer processes these emergent factors through various ways (such as weighted average, smoothing processing, etc.), ensures the stability of the evaluation results, and thereby enhances the robustness. When encountering sudden situations such as the addition or reduction of supporting facilities around a certain housing, the strategy layer can quickly adjust the corresponding parameters or adopt emergency strategies to timely correct the evaluation results. This rapid response mechanism ensures that the evaluation system can adapt to the dynamically changing market environment, thereby improving the flexibility.

[0024] The above is the whole process analysis of the geographic object value evaluation method described in the embodiment of the present application. The geographic object value evaluation method described in the embodiment of the present application comprises: determining a target geographic object; obtaining attribute data of the target geographic object, determining a preliminary value evaluation result of the target geographic object according to the attribute data, and taking the preliminary value evaluation result as a first result; determining at least one similar geographic object of the target geographic object, obtaining historical transaction prices of all similar geographic objects, and calculating a statistical value of all historical transaction prices, taking the statistical value as a second result of a similar prediction; obtaining historical transaction prices of other geographic objects in a region where the target geographic object is located at different times, determining a predicted value of an evaluation result of the current region according to a fitting curve determined based on the historical transaction prices and time, taking the predicted value as a third result, and calculating a ratio of the third result to a historical transaction price corresponding to a previous time unit on the fitting curve, taking the ratio as a trend change coefficient; determining a missing degree of the similar geographic objects according to a number of the similar geographic objects of the target geographic object and a number of transaction objects in the region where the target geographic object is located; and using the missing degree as an adjustment coefficient, using a product of the second result and the trend change coefficient as an adjustment amount, and adjusting the first result to obtain a value evaluation result of the target geographic object.

[0025] The geographic object value evaluation method described in the embodiment of the present application comprehensively considers the preliminary value evaluation result determined based on the attribute data of the target geographic object, the statistical value of the historical transaction prices of all similar geographic objects similar to the target geographic object, the predicted value of the evaluation result of the current region determined according to the fitting curve determined based on the historical transaction prices of other geographic objects in the region where the target geographic object is located at different times and time, the trend change coefficient determined based on the ratio of the predicted value to the historical transaction price corresponding to the previous time unit on the fitting curve, and the influence of the missing degree of the similar geographic objects on the value evaluation result of the target geographic object when obtaining the value evaluation result of the target geographic object, thereby improving the accuracy of the value evaluation of the target geographic object. Moreover, the data required for evaluation calculation is ordered and determined, thereby reducing the consumption of computing resources, improving the efficiency of evaluation calculation, and improving the stability of the calculation result.

[0026] The present application is described in detail through multiple embodiments and drawings.

[0027] First Embodiment The first embodiment of the present application provides a geographic object value evaluation method, which is described below in combination with Figure 3 The geographic object value evaluation method is described in detail.

[0028] Step S301: determining a target geographic object.

[0029] The step is used for determining a target geographic object. Taking a house as an example, the geographic object can be a house, an apartment, a villa, a bungalow, etc. The target geographic object refers to a geographic object that needs to be evaluated for value.

[0030] In step S302, attribute data of the target geographic object is obtained, and a preliminary value evaluation result of the target geographic object is determined according to the attribute data as a first result.

[0031] The step is used for obtaining attribute data of the target geographic object, and determining a preliminary value evaluation result of the target geographic object according to the attribute data as a first result.

[0032] In the embodiment of the present application, the attribute data includes feature data of the geographic object, set region data to which the geographic object belongs, geographic information of the geographic object, and peripheral accessory information data. The geographic information of the geographic object is relationship information between the target geographic object and a peripheral associated geographic object, the peripheral associated geographic object is a geographic object that has an associated influence on a user in the target geographic object and is within a first preset distance range from the target geographic object. The peripheral accessory information data is relationship data between a region where the target geographic object is located and a neighbor region, and the neighbor region is a region within a second preset distance range from the region where the target geographic object is located.

[0033] The feature data of the geographic object can be numerical information of the geographic object or text information of the geographic object. In the embodiment of the present application, the target geographic object is a geographic object to be evaluated for value. Taking a house as an example, the geographic object can be various types of houses, such as a commercial house, a residential house, an apartment, a villa, a bungalow, a garage, a warehouse, etc. The numerical information can be related attribute information of the house and the community where the house is located. The house information can include building year, area, floor, room, orientation, etc. The information of the community where the house is located can include green rate, plot ratio, floor area ratio, volume fraction, water, electricity, fuel cost, etc. The text information mainly includes description information of the house, which can be house introduction, community introduction, peripheral supporting facilities, traffic conditions, etc.

[0034] It should be noted that the building year, area, floor and room type of the house source can directly reflect the basic situation of the house source, and are important factors for evaluating the value of the house source. Whether the supporting facilities around the community are perfect, whether the traffic around the community is convenient, and whether the community is located in a commercial prosperous area are important factors affecting the house source evaluation result.

[0035] The set region to which the geographic object belongs refers to a community where the target geographic object (for example, a house) is located, or a sub-community of the community, for example, some house sources belong to a first area, a second area, or an east area, a south area, etc. The set region data to which the geographic object belongs refers to characteristic data of the community where the target geographic object is located, or characteristic data of the sub-community of the community. The characteristic data of the community where the geographic object is located can also include quantifiable numerical information and descriptive text information. The set region data to which the geographic object belongs can refer to data of the community where the house source is located. The community refers to a large piece of residential housing with a relatively independent living environment in a certain area of a city, and is also provided with a complete set of life service facilities.

[0036] The geographic information of the geographic object can refer to the splicing of the house source and surrounding subway stations, hospitals and the like, or the convenience degree information of the supporting entity to the house source obtained through calculation. In the embodiment of the application, the surrounding associated geographic object is a geographic object having an associated influence on the user in the target geographic object and having a distance within a first preset distance range from the target geographic object. The surrounding associated geographic object can be supporting entity information, which refers to surrounding geographic factors that have a significant influence on the value of the house, such as schools, subways, hospitals, parks, supermarkets, shopping malls or other important entities, etc. In specific implementation, an entity list can be preset to determine the influence degree of the entity in the entity list on the house source, or the house source data and the entity data can be input into a preset model to predict the influence degree of the entity on the house source through the model, and then determine whether the entity is an associated entity of the house source. It should be noted that the surrounding associated geographic object refers to the surrounding associated geographic object of the house source, which can be an entity outside the community where the house source is located, or an entity inside the community where the house source is located, such as a convenience store, a school, etc. inside the community, a convenience store and a fast food restaurant, etc. in a commercial building. The surrounding not only refers to outside the community or outside the commercial building, but also can refer to inside the community or inside the commercial building. The surrounding accessory information data can be relationship data between the community where the target geographic object is located and the surrounding adjacent communities.

[0037] In the embodiments of the present application, the preliminary value evaluation result of the target geographic object can be determined according to the attribute data as the first result. When determining the preliminary value evaluation result of the target geographic object according to the attribute data, specifically, the feature data of the geographic object, the set region data to which the geographic object belongs, the geographic information of the geographic object and the peripheral auxiliary information data are input into a preset geographic object value evaluation model to determine the preliminary value evaluation result of the target geographic object, i.e., the first result. .

[0038] The geographic object value evaluation model is a regression model obtained through training. The geographic object value evaluation model is trained in the following manner: sample data of geographic object set data composed of a plurality of geographic objects for which value evaluation has been completed is determined, the sample data including feature data samples of geographic objects in the geographic object set data, set region data samples to which the geographic objects belong, geographic information samples of the geographic objects, peripheral auxiliary information data samples and actual result samples of the geographic objects in the geographic object set data for which value evaluation has been completed; each feature data sample of the geographic objects, each set region data sample of the geographic objects, each geographic information sample of the geographic objects, each peripheral auxiliary information data sample and each actual result sample are taken as a set of training sample pairs, the training sample pairs being used to train the geographic object value evaluation model; an initial geographic object value evaluation model is determined; the feature data samples of the geographic objects, the set region data samples of the geographic objects, the geographic information samples of the geographic objects and the peripheral auxiliary information data samples are input into the initial geographic object value evaluation model to determine initial value evaluation results of the geographic objects; if loss data between the initial value evaluation results and the actual result samples satisfies an acceptable loss condition, the initial geographic object value evaluation model is determined as the geographic object value evaluation model, otherwise, model parameters in the initial geographic object value evaluation model are adjusted until the loss data between the value evaluation results obtained by the geographic object value evaluation model after the model parameters are adjusted and the actual result samples satisfies the acceptable loss condition, and the geographic object value evaluation model after the model parameters are adjusted is determined as the geographic object value evaluation model.

[0039] The above is a training method of the geographic object value evaluation model. In actual implementation, the LightGMB regression model method can also be used for model training according to the sample data of the house source. LightGBM is a fast, distributed, and high-performance gradient boosting framework based on decision tree algorithm. It is mainly used for sorting, classification, regression, and many other machine learning tasks. The main advantages of LightGBM include faster training speed, higher efficiency, lower memory occupation, and higher accuracy. In order to improve the fitting ability of the model to abnormal data, the loss function can be trained using Huber Loss, that is, wherein, is the actual result, is the estimated result, = , is a hyperparameter that determines the point at which the loss function changes from squared error to absolute error.

[0040] The above is the process of determining the preliminary value evaluation result of the target geographic object according to the attribute data of the target geographic object, as the first result. The following describes the process of determining at least one similar geographic object of the target geographic object, obtaining the historical transaction prices of all similar geographic objects, and calculating the statistical value of all historical transaction prices as the second result of the similar prediction.

[0041] Step S303: Determine at least one similar geographic object of the target geographic object, obtain the historical transaction prices of all similar geographic objects, and calculate the statistical value of all historical transaction prices as the second result of the similar prediction.

[0042] This step is used to determine the similar geographic object of the target geographic object. The similar geographic object refers to a geographic object with similar characteristic data to the target geographic object. Obtaining the historical transaction prices of all similar geographic objects and calculating the statistical value of all historical transaction prices can significantly improve the accuracy and reliability of the value prediction result of the target geographic object.

[0043] In the field of house price estimation, it is crucial to accurately estimate the price of a house. In order to improve the accuracy of the prediction, various methods can be used to reference the characteristics of similar houses to obtain the estimated value of similar houses. One method is to determine the similar geographical objects using predefined rules. The predefined rules include: all or part of other geographical objects in the designated area where the target geographical object belongs are considered as similar geographical objects; other geographical objects in the designated area and / or adjacent designated area with the same or similar area as the target geographical object are considered as similar geographical objects; other geographical objects in the designated area and / or adjacent designated area with the same or similar floor as the target geographical object are considered as similar geographical objects; other geographical objects in the designated area with the same or adjacent building as the target geographical object are considered as similar geographical objects. For example, houses in the same community: houses in the same community usually have little difference in geographical location and environment, so houses in the same community are preferred as similar houses for comparison. Houses with similar areas: houses are classified by area, and houses with similar areas to the target house are usually selected, such as houses within ±10% of the target house. Houses with similar floors: the height of the floor can affect the price, so houses with similar floors to the target house are selected for comparison, such as houses within ±2 floors. Houses with similar buildings: in the same community, prices may also differ between different buildings, so by selecting houses in the same or similar buildings, the error of the estimated result can be further reduced.

[0044] Another method is to determine the similar geographical objects of the target geographical object through clustering, which specifically includes: determining other geographical objects in the area where the target geographical object is located or in the adjacent area of the area; obtaining attribute data of the other geographical objects; clustering the other geographical objects according to attribute categories as clustering features based on the attribute data; and determining similar geographical objects of the target geographical object from different clusters. The determination of similar geographical objects of the target geographical object from different clusters includes: determining the similarity of other geographical objects in different clusters to the target geographical object, and considering other geographical objects with a similarity exceeding a set threshold as similar geographical objects of the target geographical object.

[0045] The clustering method is to group the housing sources according to multiple dimensions through unsupervised learning technology in data mining, and to estimate the price by analyzing similar groups. Through geographic location, historical performance price, housing layout, etc., unsupervised clustering is performed, and similar housing information is found through multi-path recall. The main process is as follows: 1) feature selection: in order to cluster, first, select appropriate features, such as geographic location (latitude and longitude coordinates), historical transaction price, housing layout (such as the number of rooms, area), etc. 2) data standardization: in order to ensure the relative influence between different features, the data can be standardized to make each feature in the same order of magnitude. 3) select clustering algorithm: use K-means, DBSCAN, etc. Clustering algorithm to cluster housing data, the similarity of housing in each class is high. 4) multi-path recall: clustering from multiple angles (geographic location, price, layout), each clustering method can find similar housing, and through comprehensive evaluation of similar housing under different angles, more accurate housing information is obtained.

[0046] Another way is to use a preset model to encode the feature data of the geographic object to obtain an encoded feature vector; and to cluster the feature vectors based on a preset clustering condition to obtain a set of similar geographic objects. Specifically, a deep learning model is used to learn high-level features of the housing source through unsupervised or semi-supervised methods, and then clustering and similar matching are performed according to the learned representation. The main process is as follows: 1) data preparation: the data includes text features (housing description), numerical features (area, price, etc.) and picture features (housing pictures). 2) model selection: use a deep learning model, such as Autoencoder or Variational Autoencoder (VAE) for unsupervised learning, to learn high-level features from the data. 3) representation learning: use the model to learn the representation of the housing features, and map the housing data to a high-dimensional feature space, at this time each housing has a higher-dimensional feature representation. 4) similar housing clustering: clustering in the high-dimensional feature space, so that similar houses are aggregated into a class, and through these similar housing information, the price estimation of the housing to be evaluated is performed. 5) model optimization: semi-supervised learning method can be used to train the model with a part of the data with known labels to improve the representation ability of the model.

[0047] In specific implementation, the similarity between the other geographic object and the target geographic object is calculated by the following method: obtaining a vector representation of the attribute data of the target geographic object as a first vector A according to the attribute data of the target geographic object; obtaining a vector representation of the attribute data of any other geographic object as a second vector B; calculating the cosine similarity of the first vector and the second vector as the similarity between the other geographic object and the target geographic object , that is In determining the similarity according to the location of the geographic object, the similarity of the other geographic object to the target geographic object is calculated by determining the geographic distance between the target geographic object and the other geographic object according to the geographic location information of the target geographic object and the geographic location information of the other geographic object, and taking the geographic distance as the similarity of the other geographic object to the target geographic object. The geographic location information of the target geographic object can be longitude and latitude data, and the geographic distance can be Euclidean distance or Manhattan distance. In determining the similarity according to the house type of the geographic object, the similarity of the other geographic object to the target geographic object is calculated by first converting the house type of the target geographic object into a numerical feature representation as a third vector, converting the house type data of any other geographic object into a numerical feature representation as a fourth vector, and calculating the cosine similarity of the third vector and the fourth vector as the similarity of the other geographic object to the target geographic object. In determining the similarity according to the area of the geographic object, the similarity of the other geographic object to the target geographic object is calculated by normalizing the area of the other geographic object, and taking the normalized result as the similarity of the other geographic object to the target geographic object. For example, the absolute values of the normalized results of the areas of the target geographic object and the other geographic object are subtracted, and the result of the area similarity is closer to zero, the closer the areas of the housing sources are, and the higher the similarity of the housing sources is. The normalization of the area is as follows: ​​​​​​​​​​ It should be noted that the above is a way of determining similarity by using attribute data, location information, house type, and area between geographic objects. The final house source similarity can also be obtained by using a weighted average method by combining multiple feature similarities In actual implementation, the corresponding similarity calculation method can be selected according to actual needs. The following is a calculation method of comprehensive similarity , wherein is the location similarity, is the weight of the location similarity, is the house type similarity, is the weight of the house type similarity, is the area similarity, is the weight of the area similarity, can be obtained by model training: The above is the process of obtaining the similarity of other geographic objects and the target geographic object and the similar geographic object set. In actual implementation, a suitable method or a combination of several methods can be selected to obtain the similar geographic object set according to actual needs, so as to improve the accuracy of house price prediction. By combining the results of multiple methods for comprehensive evaluation, the accuracy and reliability of price estimation can be significantly improved. After determining at least one similar geographic object of the target geographic object, the historical transaction prices of all similar geographic objects are obtained, and the statistical value of all historical transaction prices is calculated as a second result of similar prediction. The calculation of the statistical value of all historical transaction prices includes: determining the similarity representing the correlation degree between each similar geographic object and the target geographic object, and calculating the statistical value of all historical transaction prices by weighted average of all historical transaction prices using the similarity as a weight. The calculation of the statistical value of all historical transaction prices includes: determining the similarity representing the correlation degree between each similar geographic object and the target geographic object, and calculating the statistical value of all historical transaction prices by weighted average of all historical transaction prices using the similarity as a weight.

[0048] After obtaining the similar geographic object set , further obtain the similarity data between each of the similar geographic objects in the similar geographic object set and the target geographic object , which can be any of the multiple individual similarities calculated above, or the comprehensive similarity calculated above, and the historical transaction price of each of the similar geographic objects in the similar geographic object set ; based on the similarity data between each of the similar geographic objects and the target geographic object and the historical transaction price of each of the similar geographic objects , the statistical value of all historical transaction prices is calculated by weighted average of all historical transaction prices using the similarity as weight , i.e. wherein, is the similarity data, and in actual calculation, if the similarity is calculated according to the location similarity, the similarity is calculated according to the above value, if the similarity is calculated according to the house type similarity, the similarity is calculated according to the above value, if the similarity is calculated according to the comprehensive similarity, the similarity is calculated according to the above value, if the similarity is calculated according to the area or price similarity, the similarity is calculated according to the above area similarity or price similarity value, is the historical transaction price of the similar house source, is the similar house source set. Finally, the statistical value of all historical transaction prices is calculated by arithmetic average or median of all historical transaction prices.

[0049] Step S304: obtaining the historical transaction prices of other geographical objects in the region where the target geographical object is located at different times, determining the predicted value of the evaluation result of the current region according to the fitting curve determined according to the relationship between the historical transaction prices and time, taking the predicted value as the third result, calculating the ratio of the third result to the historical transaction price corresponding to the previous time unit on the fitting curve as the trend change coefficient.

[0050] In obtaining the trend change coefficient, specifically, the historical transaction prices of other geographical objects in the region where the target geographical object is located at different times are obtained first, wherein the region where the target geographical object is located can be within a preset community range, can be a joint region of surrounding communities, can be a region within a preset distance range centered on the target geographical object, can be a district information to which the target geographical object belongs, or can be a city to which the target geographical object belongs, and in actual implementation, it can be set according to actual needs. After obtaining the historical transaction prices of other geographical objects in the region where the target geographical object is located at different times, the predicted value of the evaluation result of the current region can be determined according to the fitting curve determined according to the relationship between the historical transaction prices and time, taking the predicted value as the third result, i.e. .

[0051] In actual implementation, the house prices of different regions can be aggregated according to the month dimension to construct a time series of region, month and average price dimensions, so as to obtain the house price trend data of the region. Please refer to Figure 4 , Figure 4This is a schematic diagram of the fitting curve in a geographic object valuation method provided in the first embodiment of this application. According to the fitting curve, the housing price trend of the region can be obtained. For example, based on historical transaction data (such as average price) before January 2024, the estimated housing price data for March 2024 and April 2024 can be predicted. Figure 4 (In the middle circle). It should be noted that the embodiments of this application use the Prophet model. The basic principle of Prophet is to decompose the time series into a trend term, a seasonal term, and a holiday term, and then use an additive model for prediction based on this. Specifically, it is assumed that the time series consists of the following three parts: trend term... : Describes the long-term trend of the time series, using a piecewise linear model with an autoregressive term and a seasonal term for fitting; the seasonal term : Describes the periodic changes of a time series, using Fourier series for fitting; holiday term : Describes abnormal events (such as holidays) in a time series, represented by a custom binary variable; Error term: Represents variations that the model failed to capture; it is usually assumed to be normally distributed. Adding the trend term, seasonal term, and holiday term yields the predicted value of the time series. Right now In practice, in order to obtain the trend change coefficient In determining the predicted value of the current assessment results for this region. Then, obtain the historical transaction prices corresponding to the previous time unit (e.g., the previous month) on the fitted curve. Trend change coefficient The third result Historical transaction prices corresponding to the previous time unit on the fitted curve The ratio, i.e. It should be noted that the previous time unit on the fitted curve is determined based on the time unit used in the fitting curve according to the relationship between historical transaction prices and time. For example, if the time unit used in the fitting curve is year, then the previous time unit is the previous year; if the time unit used in the fitting curve is month, then the previous time unit is the previous month; and if the time unit used in the fitting curve is quarter, then the previous time unit is the previous quarter.

[0052] Step S305: Determine the missingness of the similar geographic objects based on the number of similar geographic objects of the target geographic object and the number of transactions in the region where the target geographic object is located.

[0053] This step is used to determine the degree of missing similar geographic objects.

[0054] In the determination of the missing degree data of the similar geographical object, firstly, the number of similar geographical objects of the target geographical object and the number of transaction objects in the region where the target geographical object is located are obtained, then, the difference between the number of similar geographical objects of the target geographical object and the number of transaction objects in the region where the target geographical object is located is obtained, and finally, the ratio of the difference to the number of transaction objects in the region where the target geographical object is located is obtained as the missing degree of the similar geographical object. For example, the number of similar geographical objects of the target geographical object is 15, and the number of transaction objects in the region where the target geographical object is located is 20, and then the missing degree is : It should be noted that if the number of similar geographical objects of the target geographical object is 30 and the number of transaction objects in the region where the target geographical object is located is 20, since the number of similar geographical objects of the target geographical object exceeds the number of transaction objects in the region where the target geographical object is located, the missing degree is 0, and if the number of transaction objects in the region where the target geographical object is located is 0, the missing degree is 1.

[0055] Step S306: Adjusting the first result by using the missing degree as an adjustment coefficient and the product of the second result and the trend change coefficient as an adjustment amount to obtain the value evaluation result of the target geographical object.

[0056] This step is used to obtain the value evaluation result of the target geographical object, which combines the preliminary value evaluation result of the target geographical object, the second result of the similarity prediction, the trend change coefficient and the missing degree of the similar geographical object, so that the finally obtained value evaluation result is more accurate.

[0057] In the embodiment of the present application, the use of the missing degree as an adjustment coefficient and the product of the second result and the trend change coefficient as an adjustment amount to adjust the first result to obtain the value evaluation result of the target geographical object comprises: obtaining a negative missing degree corresponding to the first result according to the full value corresponding to the missing degree and the missing degree; taking the negative missing degree as the weight of the first result and taking the missing degree as the weight of the second result to obtain the weighted sum data of the first result and the second result; taking the trend change coefficient as the adjustment factor data of the weighted sum data, using the trend change coefficient to adjust the weighted sum data to obtain the value evaluation result of the target geographical object.

[0058] It should be noted that in the obtaining of the value evaluation result of the target geographical object according to the first result , the second result and the trend change coefficient ​​ When, specifically, the missing degree of the similar geographical object is determined After that, it is assumed The range of the missing degree is 0-1, the first result The corresponding negative missing degree is 1- The negative missing degree 1- is taken as the weight of the first result The missing degree is taken as the weight of the second result The weighted sum data of the first result and the second result is obtained, and then the trend change coefficient is taken as the adjustment factor data of the weighted sum data, the weighted sum data is adjusted using the trend change coefficient to obtain the value evaluation result of the target geographical object , that is It should be noted that when the missing degree tends to 0, the house source estimation result tends to the first result , and when the missing degree tends to 1, the house source estimation result tends to the second result . That is, when the number of transaction objects in the area where the target geographical object is located is relatively rich, the result obtained by using the geographical object value evaluation model can be used, and when the number of transaction objects in the area where the target geographical object is located is relatively sparse, the transaction result of the similar geographical object can be referred to.

[0059] The above is the process of obtaining the value evaluation result of the target geographical object by using the single house source prediction module, the similar house source price prediction module, and the house price trend estimation module. In the similar house source price prediction module, the clustering information part mainly provides similar house sources and similar community information for new house sources and transaction sparse communities, and increases the robustness of the overall model. In specific implementation, a first additional result of the house source to be evaluated can also be obtained according to the category data and the attribute data of the house source to be evaluated, and a second additional result of the house source to be evaluated can also be obtained according to the city data and the attribute data to which the house source to be evaluated belongs. Finally, the value evaluation result, the first additional result, and the second additional result are fused to obtain a total evaluation result of the house source to be evaluated, so that the final house source value estimation result is more accurate. Specifically, the category data and the city data to which the target geographical object belongs are obtained; the evaluation result of the target geographical object is determined according to the category data and the attribute data, as a first additional result; the evaluation result of the target geographical object is determined according to the city data and the attribute data, as a second additional result; the value evaluation result, the first additional result, and the second additional result are fused to obtain a total evaluation result of the target geographical object.

[0060] The determining of the evaluation result of the target geographic object according to the category data and the attribute data as the first additional result comprises: determining a category data obtaining sub-model; and determining the evaluation result of the target geographic object based on the category data obtaining sub-model and the attribute data as the first additional result. The determining of the evaluation result of the target geographic object according to the city data and the attribute data as the second additional result comprises: determining a city data obtaining sub-model; and determining the evaluation result of the target geographic object based on the city data obtaining sub-model and the attribute data as the second additional result.

[0061] It should be noted that the category data of the target geographic object can represent the type of the target geographic object, such as a residence, an apartment, a villa, a foreign house, etc. For example, if the type of the target geographic object is an apartment, considering the special attributes of the apartment type, an apartment sub-model can be selected, and an apartment value evaluation sub-model can be constructed based on the historical transaction records of the apartment by using the attribute data such as the geographic location of the apartment, the area of the apartment, the number of rooms, the construction year, the supporting facilities, the rent price, the water and electricity property fee, etc. The training method of the apartment value evaluation sub-model is similar to the training method of the above-mentioned geographic object value evaluation model, and can be referred to the training method of the above-mentioned geographic object value evaluation model, which will not be described in detail here.

[0062] The above is one training method of the apartment value evaluation sub-model, and in specific implementation, the GBDT model can also be used to train the apartment based on the sample data of the apartment. The GBDT model (Gradient Boosting Decision Tree) is a kind of enhanced learning algorithm, which is commonly used for regression and classification problems, and it improves the accuracy of the model by combining multiple weak learners (usually decision trees) into a strong learner. The GBDT model mainly constructs decision trees in an iterative manner, and each construction tries to correct the error of the previous model, and the basic flow is as follows: 1, initial model: first, use a simple model (usually the average value) to initialize the prediction, for the regression problem, the initial output is the mean value of the target value of the training data. 2, calculate the residual error: use the current model to predict the training data, calculate the error (residual error) of each sample, the residual error is the difference between the true value and the predicted value. 3, build a new tree: build a new decision tree based on the residual error, the new tree aims to minimize the loss function of the current model, specifically, it will fit the residual error calculated in the last step. 4, update the model: add the prediction result of the new tree to the prediction value of the current model, usually through a learning rate (shrinkage) to control the influence of each tree on the final prediction result. 5, iteration: repeat steps 2 to 4 until the number of trees reaches the preset number, or the residual error does not change significantly.

[0063] In implementation, in addition to the apartment value evaluation sub-model, a villa value evaluation sub-model, a garage value evaluation sub-model, etc. can also be constructed. For the villa value evaluation sub-model, considering the special attributes of villa types, a model can be constructed separately for villa types, considering the address location, villa area, total floor, number of rooms, architectural style, construction year, volume rate, property fee, etc. of the villa, and based on the historical transaction records of the villa, a villa value evaluation sub-model is constructed. For the training of the villa value evaluation sub-model, the training method of the above-mentioned apartment value evaluation sub-model can also be used for training, and the GBDT model can also be used for training. For the garage value evaluation sub-model, the garage value evaluation sub-model can be used for the house containing the parking space, and the value of the parking space is estimated separately. The price of the parking space, the size of the parking space, the type of the parking space, the rent, etc. are comprehensively considered, and the value of the parking space is estimated.

[0064] It should be noted that the city data of the target geographic object can be obtained through the city sub-model and the regional sub-model. For the city sub-model, considering the different development levels of cities, such as the more active transaction of second-hand houses in first-tier cities and the higher unit price, and the sparse transaction of second-hand houses in third and fourth-tier cities and the lower unit price, the city sub-model is constructed based on the development level of the city. The training method of the city sub-model is similar to the training method of the above-mentioned sub-model, which will not be described in detail here. For the regional sub-model, considering the different development levels of different regions, such as the higher flow of houses in eastern cities than in western cities, the regional sub-model is constructed based on the division of addresses in the country, and the regional sub-model is constructed separately according to different regions. The training method of the regional sub-model is similar to the training method of the above-mentioned sub-model, which will not be described in detail here.

[0065] It should be noted that the geographic object value evaluation model as the evaluation main model is the core model, and the evaluation main model integrates multiple key factors such as historical transactions, similar houses, market trends, etc. The purpose is to provide preliminary and comprehensive valuation reference. The sub-model carefully considers the unique attributes of specific types of houses, and in the complex real estate market, a single model is difficult to adapt to all situations. Therefore, multiple sub-models are customized for modeling according to different residential properties and geographic locations. For example, the apartment value evaluation sub-model may specially consider the special attributes of the apartment, and the city sub-model pays more attention to the city level of the house. Although each sub-model may be similar in structure, there will be significant differences in the processing of specific factors to ensure that the prediction of specific house types is more accurate.

[0066] In the embodiments of the present application, the value evaluation result is fused with the first additional result and the second additional result to obtain a total evaluation result of the target geographic object, including: obtaining a first weight of the value evaluation result; obtaining a second weight of the first additional result and the second additional result; and obtaining a total evaluation result of the target geographic object based on the value evaluation result, the first weight, and the first additional result, the second additional result, and the second weight.

[0067] In specific implementation, the value evaluation result can be a main result, and the first additional result and the second additional result can be auxiliary results. The total evaluation result of the target geographic object is obtained based on the main result and the auxiliary results. The total evaluation result of the target geographic object is obtained based on the main result and the auxiliary results. In specific implementation, the total evaluation result of the target geographic object is obtained based on the type of the house source and the city to which the house source belongs. Different weights are learned for different sub-models based on training data by using an LR model, and finally the final model prediction result is obtained by weighting. Specifically, the weighted sum data of the main result and the auxiliary results is obtained as the total evaluation result of the target geographic object based on the first weight of the main result and the second weight of the auxiliary results. It should be noted that when the total evaluation result of the target geographic object is obtained based on the main result and the auxiliary results, The auxiliary result can be the sum of a plurality of auxiliary results.

[0068] The above is the overall process of obtaining the total evaluation result of the target geographic object. In specific implementation, after obtaining the value evaluation result or the total evaluation result of the target geographic object, a dynamic correction strategy is used to correct the value evaluation result or the total evaluation result of the target geographic object, including: determining a favorable or unfavorable event that affects the value evaluation of the target geographic object; determining the influence degree of the favorable or unfavorable event on the current value evaluation as a correction coefficient; and correcting the value evaluation result or the total evaluation result of the target geographic object by using the correction coefficient.

[0069] ​​​​​​​​​​​​It should be noted that after obtaining the total evaluation result of the target geographic object, the value evaluation result or the total evaluation result of the target geographic object is also corrected. Specifically, first, favorable or unfavorable events affecting the value evaluation of the target geographic object are determined, then the influence degree of the favorable or unfavorable events on the current value evaluation is determined as a correction coefficient, and finally the value evaluation result or the total evaluation result of the target geographic object is corrected by using the correction coefficient. For example, the surrounding area of the geographic position where the geographic object is located temporarily increases or decreases supporting facilities; it can also be determined based on the characteristic data of the demand side of the geographic object, such as the number of housing sources owned by the house buyer. The above situations will all affect the evaluation result of the housing source. Through the correction coefficient, the housing evaluation result can better adapt to the dynamically changing market environment.

[0070] The above is a full-process description of the geographic object value evaluation method described in the embodiments of the present application. On the one hand, the method solves the limitation of the similar house source recognition ability in the prior art. Most of the existing housing evaluation models rely only on the historical transaction data of the current community to evaluate the housing value, ignoring the comprehensive analysis of similar houses and similar communities, resulting in obvious defects in the generalization performance of the model. For example, different communities may have significant differences in housing prices even if they are close in geographic location due to differences in building quality, supporting facilities, and other factors. Therefore, relying only on the data of a single community for evaluation may ignore these important external factors. On the other hand, the method solves the problem that the existing technology is difficult to adapt to the rapid changes in market dynamics. The housing market price is affected by many complex factors such as supply and demand relationship, etc. These factors are usually difficult to predict and change rapidly. However, many existing evaluation models are constructed based on static data and cannot timely reflect the latest market dynamics, especially in the case of large market fluctuations, which may cause a large deviation between the evaluation result and the actual price. On the other hand, the method solves the problem of insufficient consideration of geographic heterogeneity. The housing prices in different cities or even different regions within the same city may have great differences. If a general evaluation model does not fully consider the geographic heterogeneity, it may lead to large errors in the evaluation result. For example, the housing prices in the city center and the suburbs may be very different, and some existing models cannot accurately capture the differences between regions, thereby affecting the accuracy of the evaluation. In contrast, the geographic object data processing method described in the embodiments of the present application comprehensively considers a single house, similar houses, and market trends. The model is not only based on the house information within a single community, but also incorporates the historical transaction data of similar houses and similar communities to enhance the robustness and accuracy of the model. In addition, by analyzing the characteristics of housing price fluctuations in different regions, the model can learn and capture the trends of housing prices, thereby improving the prediction ability of future market trends.

[0071] The geographic object value evaluation method provided in the embodiments of the present application comprises the following steps: determining a target geographic object; obtaining attribute data of the target geographic object, determining a preliminary value evaluation result of the target geographic object according to the attribute data, and taking the preliminary value evaluation result as a first result; determining at least one similar geographic object of the target geographic object, obtaining historical transaction prices of all similar geographic objects, and calculating a statistical value of all historical transaction prices, taking the statistical value as a second result of a similar prediction; obtaining historical transaction prices of other geographic objects in a region where the target geographic object is located at different times, determining a predicted value of an evaluation result of the region at present according to a fitting curve determined based on the historical transaction prices and time, taking the predicted value as a third result, and calculating a ratio of the third result to a historical transaction price corresponding to a previous time unit on the fitting curve, taking the ratio as a trend change coefficient; determining a missing degree of the similar geographic objects according to a number of the similar geographic objects of the target geographic object and a number of transaction objects in the region where the target geographic object is located; and adjusting the first result by using the missing degree as an adjustment coefficient and a product of the second result and the trend change coefficient as an adjustment amount, to obtain a value evaluation result of the target geographic object.

[0072] The geographic object value evaluation method provided in the embodiments of the present application comprehensively considers the preliminary value evaluation result determined based on the attribute data of the target geographic object, the statistical value of the historical transaction prices of all similar geographic objects similar to the target geographic object, the predicted value of the evaluation result of the region at present determined according to the fitting curve determined based on the historical transaction prices of other geographic objects in the region where the target geographic object is located at different times and time, the trend change coefficient determined based on the ratio of the predicted value to the historical transaction price corresponding to the previous time unit on the fitting curve, and the influence of the missing degree of the similar geographic objects on the value evaluation result of the target geographic object when obtaining the value evaluation result of the target geographic object, thereby improving the accuracy of the value evaluation of the target geographic object. Moreover, the data required for the evaluation calculation is ordered and determined, thereby reducing the consumption of the calculation resources, improving the efficiency of the evaluation calculation and the stability of the calculation result.

[0073] Second embodiment In the first embodiment described above, a geographic object value evaluation method is provided. Correspondingly, the second embodiment of the present application provides a geographic object value evaluation device. Since the device embodiment is basically similar to the first embodiment of the method, the description is relatively simple, and the related parts can be referred to the part of the method embodiment. The device embodiment described below is only illustrative.

[0074] Please refer to Figure 5 FIG. 1 is a schematic diagram of a geographic object value evaluation device provided in the second embodiment of the present application.

[0075] The geographic object value evaluation device 500 comprises: Unit 501 is used to determine the target geographic object; The first result evaluation unit 502 is used to obtain attribute data of the target geographic object and determine the preliminary value evaluation result of the target geographic object based on the attribute data, as the first result; The second result acquisition unit 503 is used to identify at least one similar geographic object of the target geographic object, obtain the historical transaction prices of all similar geographic objects, and calculate the statistical value of all historical transaction prices as the second result of similarity prediction. The trend change coefficient determination unit 504 is used to obtain the historical transaction prices of other geographical objects in the area where the target geographical object is located at different times, and determine the predicted value of the current evaluation result of the area based on the fitting curve determined by the relationship between the historical transaction price and time, as the third result. The ratio of the third result to the historical transaction price corresponding to the previous time unit on the fitting curve is calculated as the trend change coefficient. The missing value determination unit 505 is used to determine the missing value of the similar geographical objects based on the number of similar geographical objects of the target geographical object and the number of transaction objects in the area where the target geographical object is located; The value assessment result determination unit 506 is used to adjust the first result by using the missing degree as an adjustment coefficient and the product of the second result and the trend change coefficient as an adjustment amount, so as to obtain the value assessment result of the target geographic object.

[0076] Third Embodiment Corresponding to the method embodiments described above, the third embodiment of this application also provides an electronic device. For example... Figure 6 As shown, Figure 6 This is a schematic diagram of an electronic device provided in the third embodiment of this application. The electronic device includes: at least one processor 601, at least one communication interface 602, at least one memory 603, and at least one communication bus 604. Optionally, the communication interface 602 can be an interface for a communication module, such as the interface for a GSM module. The processor 601 may be a CPU, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The memory 603 may include high-speed RAM and may also include non-volatile memory, such as at least one disk storage device. The memory 603 stores a program, and the processor 601 calls the program stored in the memory 603 to execute the methods provided in the above embodiments of this application.

[0077] Fourth embodiment Corresponding to the method described above, the fourth embodiment of the present application further provides a computer storage medium. The computer storage medium stores a computer program, which is run by a processor to execute the method provided in the above embodiments of the present application.

[0078] The method for evaluating the value of a geographic object provided in the above embodiments can be applied to various application scenarios, such as information prediction, information recommendation, and various information evaluation scenarios. The following mainly takes information recommendation as an example for illustration.

[0079] Figure 7 is a flowchart of an information recommendation method provided by an embodiment of the present application. As shown in Figure 7 , the method comprises the following steps: S701: Obtain portrait data and historical behavior data of a user, wherein the portrait data comprises a target geographic object associated with the user, attribute data of the target geographic object reflects preliminary value information of the target geographic object, and the historical behavior data reflects a dynamic interest intention of the user; S702: Determine similar geographic objects of the target geographic object and a quantity of the similar geographic objects according to the attribute data of the target geographic object, wherein the attribute data of the target geographic object at least comprises a target region to which the target geographic object belongs; S703: Generate a value trend change coefficient of the target region according to a fitting curve of historical transaction prices and time relationship of other geographic objects in the target region; S704: Determine a missing degree of the similar geographic objects according to the quantity of the similar geographic objects and a quantity of transaction objects in the target region; S705: Adjust the preliminary value information by using the missing degree as an adjustment coefficient and a product of a statistical value of historical transaction prices of the similar geographic objects and the trend change coefficient as an adjustment amount, to obtain target value information, wherein the target value information reflects a steady-state demand potential of the user; S706: Jointly recommend information to the user from two dimensions of the dynamic interest intention and the steady-state demand potential according to the historical behavior data and the target value information of the target geographic object.

[0080] In the embodiment, portrait data and historical behavior data of a user are obtained, and the portrait data comprises a target geographic object associated with the user. For example, the target geographic object is a house source, and the target geographic object can be a residence, an apartment, a villa, a bungalow, etc. The historical behavior data comprises but is not limited to user click, browse, search, purchase records, etc.

[0081] Further, in the embodiment, the similar geographic object and the number of the similar geographic object of the target geographic object are determined according to attribute data of the target geographic object, and the attribute data of the target geographic object at least includes a target area to which the target geographic object belongs. Optionally, the attribute data includes a region attribute and an area attribute, and the region attribute identifies that the target geographic object belongs to the target area.

[0082] In the embodiment, the target geographic object is taken as a house for example, and the attribute data of the target geographic object can be information of a community where the house is located. Further, the clustering information is obtained through the clustering model, and the similar geographic object and the number of the similar geographic object similar to the attribute data of the target geographic object are obtained.

[0083] Further, in the embodiment, the missing degree of the similar geographic object is determined according to the number of the similar geographic object and the number of the transaction object of the target area. When the missing degree data of the similar geographic object is determined, the number of the similar geographic object of the target geographic object is subtracted from the number of the transaction object of the target area, and the ratio of the obtained difference value to the number of the transaction object of the target area is taken as the missing degree of the similar geographic object.

[0084] Further, the missing degree is used as an adjustment coefficient, the product of the statistical value of the historical transaction price of the similar geographic object and the trend change coefficient is taken as an adjustment amount, the preliminary value information is adjusted to obtain the target value information, and the target value information reflects the steady-state demand potential of the user. Since the target geographic object has strong relevance with the user, the target geographic object can reflect the long-term and steady-state characteristics of the user, such as the living level, the economic level and the life circle of the user, and therefore the information can be used to represent the steady-state demand potential of the user.

[0085] In the embodiment, the user is jointly recommended information from two dimensions of the dynamic interest intention and the steady-state demand potential according to the historical behavior data and the target value information of the target geographic object.

[0086] In the embodiment, the recent interest change trend of the user is identified in the dimension of the dynamic interest intention, the long-term geographic object value preference of the user is identified in the steady-state dimension, and the joint information recommendation is performed by fusing information of the two dimensions, so as to generate the personalized recommendation result.

[0087] In the embodiments of the present application, the attribute data of the target geographic object is jointly modeled with the historical behavior data to form a collaborative recommendation link from behavior driving to value driving; the similar geographic objects and the number thereof are determined according to the attribute data of the target geographic object; the value trend change coefficient is generated based on the fitting curve of the historical transaction price and time relationship in the target region; the missing degree of the similar objects is calculated in combination with the number of similar objects and the number of transaction objects in the region; the preliminary value information is adjusted by taking the missing degree as the adjustment coefficient and taking the statistical value of the historical transaction price of the similar objects and the value trend change coefficient as the adjustment amount to obtain the target value information reflecting the steady-state demand potential of the user; and the dynamic interest intention represented by the historical behavior data and the target value information are jointly used in two dimensions for information recommendation. In this way, the traditional data paradigm for recommendation is broken through, and the recommendation effect of the recommendation system is greatly improved.

[0088] In addition to the above information recommendation method, the embodiments of the present application also provide another information recommendation method depending on the address object associated with the user. Figure 8 is a flowchart of another information recommendation method provided by the embodiments of the present application. As shown in Figure 8 the method comprises the following steps: S801: obtaining the portrait data and the historical behavior data of the user, the portrait data including the identification information of the target geographic object associated with the user; S802: obtaining the attribute data of the target geographic object from the target library according to the identification information, the target library being obtained by using the geographic object description information in the third-party library to perform data enhancement on the geographic object description information in the ontology library based on the geographic semantic features, distance features and geographic context features of the geographic object description information, the geographic object description information including the attribute data of the corresponding geographic object; S803: predicting the steady-state demand potential information of the user according to the attribute data of the target geographic object, and predicting the dynamic interest intention information of the user according to the historical behavior data; S804: jointly recommending information to the user according to the steady-state demand potential information and the dynamic interest intention information of the user in combination with the candidate information objects in the target application scenario.

[0089] In the embodiment, the portrait data is used to describe the characteristics, preferences and attribute information of the user. The obtained portrait data includes identification information of a target geographical object associated with the user. For example, when the target geographical object is a house source, the target geographical object can include but is not limited to: a residential geographical object such as a community, a residential area, a building; a public facility geographical object such as a school, a subway station, a bus station, a hospital; a commercial and service geographical object such as a commercial district, a shopping center, a commercial building, an office building; a geographical object divided according to a coverage area such as a street coverage area, a regional center, a district. In the embodiment, the target geographical object associated with the user refers to a geographical object that the user is interested in or related to. The target geographical object has corresponding identification information.

[0090] In the embodiment, the historical behavior data refers to the behavior records of the user generated on the target platform in chronological order. The target platform includes but is not limited to an e-commerce platform, a social platform, a payment platform and a logistics platform, etc. The historical behavior data includes but is not limited to user click, browse, search and purchase records, etc.

[0091] Further, according to the identification information of the target geographical object, attribute data of the target geographical object is obtained from a target library. The target library is obtained by using geographical object description information in a third-party library to enhance geographical object description information in an ontology library based on geographical semantic features, distance features and geographical context features of the geographical object description information. The geographical object description information includes attribute data of the corresponding geographical object. The geographical semantic features, distance features and geographical context features are obtained by extracting features from name and address semantics, spatial position and surrounding geographical environment for any geographical object description information.

[0092] In the embodiment, the attribute data of the geographical object includes feature data of the geographical object, set region data to which the geographical object belongs, geographical information of the geographical object and surrounding accessory information data. The geographical information of the geographical object is relationship information between the target geographical object and surrounding associated geographical objects, and the surrounding associated geographical objects are geographical objects having an associated influence on the user in the target geographical object and having a distance from the target geographical object within a first preset distance range. The surrounding accessory information data is relationship data between a region where the target geographical object is located and a neighbor region, and the neighbor region is a region having a distance from the region where the target geographical object is located within a second preset distance range.

[0093] In view of the deep dependence of the traditional behavior data system on dynamic behavior sequences and static basic portraits, the demand cognition ability is limited to the existing data paradigm, and it is difficult to break through the cognitive boundaries formed by a single behavior dimension and fixed portrait features. In the embodiment, on the one hand, by introducing the attribute data of the target geographic object, a steady-state demand inference path different from the traditional behavior data system is constructed. The attribute data of the target geographic object reflects the user's relatively stable demand or potential ability to some extent. Generally speaking, the value of the user's community or residence can reflect the user's consumption level or purchasing power. The higher the value of the community or residence, the stronger the user's consumption level or purchasing power. On the other hand, a dynamic demand inference path is constructed by using user historical behavior data. The user historical behavior data reflects the user's dynamic or short-term interest to some extent, so as to obtain the user's demand information in the "long-term steady state" and "short-term dynamic" time scales. Details are described below.

[0094] Specifically, in the embodiment, according to the attribute data of the target geographic object, the steady-state demand potential information of the user is predicted, and the long-term demand tendency of the user is predicted. At the same time, the historical behavior data of the user has time sequence characteristics, and can present the demand change trend of the user in the recent time window. Based on this, the dynamic interest intention information of the user can be predicted by using the historical behavior data, so as to identify the demand change trend of the user in the short-term time window.

[0095] Further, according to the steady-state demand potential information and the dynamic interest intention information of the user, the candidate information objects in the target application scene are combined to perform joint information recommendation for the user. In the embodiment, the above target application scene is not limited. For example, it can be e-commerce marketing, auction marketing and various information recommendation scenes.

[0096] In the embodiment, when performing joint information recommendation, the steady-state demand potential information and the dynamic interest intention information are used as the joint expression of the user demand, and are matched and analyzed with the candidate information objects in the target application scene. Therefore, by fusing the steady-state demand potential information and the dynamic interest intention information, more accurate information recommendation can be performed for the user in multiple time scales and multiple demand dimensions.

[0097] In the embodiment of the present application, by introducing the target geographic object associated with the user and its attribute data in the target library on the basis of the user portrait data and the historical behavior data, the geographic semantic features, the distance features and the geographic context features are included in the user modeling process, breaking through the data paradigm of relying only on the established behavior data and the static portrait. On this basis, the steady-state demand potential information and the dynamic interest intention information of the user are predicted according to the attribute data of the target geographic object and the historical behavior data, respectively, to depict the user demand features from the two dimensions of long-term steady-state demand and short-term dynamic interest. Further, the steady-state demand potential information and the dynamic interest intention information are combined with the candidate information objects in the target application scene for joint information recommendation, so that the recommendation result can not only fit the current interest of the user, but also maintain consistency with the long-term demand, greatly improving the recommendation effect.

[0098] In an optional embodiment, when predicting the steady-state demand potential information of the user according to the attribute data of the target geographic object, the following steps are included: determining the similar geographic objects of the target geographic object according to the attribute data of the target geographic object, the attribute data of the target geographic object at least including the target region to which the target geographic object belongs; and calling a pre-trained first information prediction model to perform information prediction according to the attribute information of the target region, the attribute data of the target geographic object and the attribute data of the similar geographic objects, to obtain the steady-state demand potential information of the user. When predicting the steady-state demand potential information of the user, the attribute data of the target geographic object and the similar geographic objects and the attribute information of the target region are introduced as inputs to the pre-trained first information prediction model, so that the first information prediction model can predict the steady-state demand potential information of the user in a richer spatial semantic and environmental context, significantly improving the accuracy and stability of the prediction result. Thus, the cognitive boundary solidification caused by excessive reliance on historical behavior sequences and static portraits is avoided, and the recommendation is no longer limited to the existing data paradigm. More comprehensive, stable and reliable steady-state demand potential information is provided for subsequent joint information recommendation, thereby significantly improving the overall recommendation performance and the user demand matching degree.

[0099] In the embodiment, the attribute data of the target geographic object at least includes the target region to which it belongs, so that the target region can be used as a basic reference to limit the range of the candidate similar objects.

[0100] In the embodiment, the determination method of the similar geographic objects is not limited. For example, according to the attribute data of the target geographic object and other geographic objects, a clustering algorithm based on attribute categories and / or a clustering algorithm based on high-dimensional feature vectors are used for clustering processing to obtain the similar geographic objects of the target geographic object. The related content of the clustering algorithm can be referred to the foregoing embodiments, which will not be described here.

[0101] Furthermore, based on the attribute information of the target area, the attribute data of the target geographic object, and the attribute data of similar geographic objects, a pre-trained first information prediction model is invoked to predict the user's steady-state demand potential information. The first information prediction model can comprehensively utilize the attribute relationships between the target geographic object and similar geographic objects to predict the user's demand tendency over a long-term time scale, thereby obtaining the user's steady-state demand potential information.

[0102] In one optional embodiment, the process of calling a pre-trained first information prediction model to predict information based on the attribute information of the target area, the attribute data of the target geographic object, and the attribute data of similar geographic objects to obtain the user's steady-state demand potential information includes: classifying the attribute information of the target area, the attribute data of the target geographic object, and the attribute data of similar geographic objects to obtain first data of multiple information modalities; inputting the first data of multiple information modalities into the first information prediction model, which includes a first global decoder, a first encoder and a first sub-decoder corresponding to each of the multiple information modalities, and performing the following operations within the first information prediction model: encoding process: inputting the first data of multiple information modalities into the first encoders corresponding to the multiple information modalities for encoding to generate first data vectors corresponding to each of the multiple information modalities; decoding process: inputting the first data vectors corresponding to each of the multiple information modalities into the corresponding first sub-decoders for local decoding, and fusing the first data vectors corresponding to each of the multiple information modalities and inputting them into the first global decoder for global decoding; prediction process: predicting the user's steady-state demand potential information based on the local value information of the target geographic object output by the multiple first sub-decoders and / or the global value information of the target geographic object output by the first global decoder. By performing local decoding on the first data vectors of each information modality in their respective first sub-decoders, local value information reflecting the local features of each information modality is obtained. Simultaneously, by fusing the first data vectors of multiple information modalities and inputting them into the first global decoder for global decoding, global value information for characterizing the comprehensive features of the multimodal model can be obtained. Therefore, the first information prediction model can utilize features from multiple information modalities simultaneously at both the local and global levels, thereby improving the accuracy and stability of predicting users' steady-state demand potential information. The resulting steady-state demand potential information is richer and more reliable, providing high-quality long-term demand data for subsequent joint information recommendation based on steady-state demand potential information and dynamic interest intent information.

[0103] In this embodiment, the data of different information modalities in the attribute information of the target area, the attribute data of the target geographic object, and the attribute data of similar geographic objects are first classified to obtain the first data of multiple information modalities.

[0104] In the embodiment, the first information prediction model comprises a first global decoder, a plurality of first encoders respectively corresponding to the plurality of information modalities, and a first decomposition decoder.

[0105] Further, the first data of the plurality of information modalities are input into the first information prediction model. In the first information prediction model, the first data of different information modalities that have been classified are respectively input into each corresponding first encoder, and through the encoding process, the first data vector corresponding to each information modality is generated.

[0106] Further, in the embodiment, the first data vector corresponding to each information modality is respectively input into the corresponding first decomposition decoder for local decoding. The first decomposition decoder is responsible for generating the corresponding local decoding result according to the first data vector of each information modality. And after the data vectors corresponding to the plurality of modalities are fused, a first fusion vector is obtained. Further, the first fusion vector is input into the global decoder for global decoding to obtain a global decoding result.

[0107] In the embodiment, the global decoding and the local decoding are relative concepts, and essentially belong to the results obtained after the model decodes the input features. Among them, the local decoding refers to the decoding process independently completed by each information modality at the level of the decomposition decoder (for example, the first decomposition decoder), which is used to restore or reconstruct the information features corresponding to the modality. The local decoding corresponds to the local decoding result. The global decoding refers to the decoding process uniformly performed by the global decoder after the local encoding results of the information modalities are fused. The global decoding obtains the global decoding result.

[0108] In the embodiment, the local decoding result output by the first decomposition decoder is the local value information, and the global decoding result output by the first global decoder is implemented as the global value information.

[0109] Further, in the embodiment, the prediction process is to predict the steady-state demand potential information of the user according to the local value information of the target geographic object output by the plurality of first decomposition decoders and / or the global value information of the target geographic object output by the first global decoder.

[0110] In the embodiment, the steady-state demand information of the user can be predicted in different ways. The following gives several implementation manners, but is not limited thereto.

[0111] In an embodiment, the final steady-state demand potential information is generated based on the first global decoder. In this embodiment, each first encoder encodes a first data vector of a different information modality, and the first data vectors are fused and globally decoded by the first global decoder to generate the final steady-state demand potential information. The first global decoder integrates data of each information modality and learns the relationship between the information modalities to output the final steady-state demand potential information.

[0112] In another embodiment, the final steady-state demand potential information is generated based on the local decoding results of the first local decoders. Each first local decoder locally decodes the first data vector corresponding thereto to generate a local decoding result corresponding to the information modality. The output of each first local decoder represents the independent contribution of the information modality to the final steady-state demand potential information. Then, the local decoding results are fused by using weighted summation to obtain the final steady-state demand potential information. The final steady-state demand potential information is the weighted fusion result of the outputs of the first local decoders.

[0113] In yet another optional embodiment, the final steady-state demand potential information is generated based on the first local decoders and the first global decoder. The first data vectors of the multiple information modalities are locally decoded by the corresponding first local decoders and are finally fused and globally decoded in the first global decoder to generate the final steady-state demand potential information.

[0114] Further, in the embodiments of the present application, the dynamic interest intention information of the user is predicted based on the historical behavior data, and the specific implementation manner of predicting the dynamic interest intention information of the user based on the historical behavior data is not limited.

[0115] In an optional embodiment, when predicting the dynamic interest intention information of the user based on the historical behavior data, the following steps are included: classifying the historical behavior data according to the information modalities presented by the historical behavior data to obtain second data of the multiple information modalities; inputting the second data of the multiple information modalities into a second information prediction model, the second information prediction model including a second global decoder, second encoders corresponding to the multiple information modalities, and second local decoders, and performing the following operations in the second information prediction model: an encoding process: inputting the second data of the multiple information modalities into the second encoders corresponding to the multiple information modalities respectively to generate second data vectors corresponding to the multiple information modalities respectively; a decoding process: inputting the second data vectors corresponding to the multiple information modalities respectively into the corresponding second local decoders to locally decode, and inputting the second data vectors corresponding to the multiple information modalities after fusion into the second global decoder to globally decode; and a prediction process: predicting the dynamic interest intention information of the user based on the local interest intention information output by the multiple second local decoders and / or the global interest intention information output by the second global decoder.

[0116] By dividing the second data into multiple information modalities according to the information modalities presented by the historical behavior data, and performing the sub-modal encoding, the sub-modal local decoding and the cross-modal global decoding respectively by using the second information prediction model including the second encoder, the second decomposition decoder and the second global decoder, the system can structure the historical behavior data of the user in three dimensions of the modal level, the local level and the global level. Compared with the prior art which usually models the historical behavior in a single sequence form, the embodiment can generate the second data vector for each information modality and obtain the local interest intent information, and then obtain the global interest intent information by combining the fused second data vector, so as to extract the dynamic interest intent information with cross-modal correlation while preserving the modal difference.

[0117] In the embodiment, first, the information modalities presented according to the historical behavior data are classified to obtain the second data of multiple information modalities. Further, the second data of multiple information modalities is input into the second information prediction model. In the second information prediction model, the second data of different information modalities classified is respectively input into each corresponding second encoder, and the second data vector corresponding to each information modality is generated by the encoding process.

[0118] Further, in the embodiment, the second data vector corresponding to each of the multiple information modalities is respectively input into the corresponding second decomposition decoder for local decoding. The second decomposition decoder is responsible for generating the corresponding local decoding result according to the second data vector of each information modality. And the data vectors corresponding to multiple modalities are fused to obtain a second fusion vector. Further, the second fusion vector is input into the second global decoder for global decoding to obtain a global decoding result.

[0119] In the embodiment, the local decoding result output by the second decomposition decoder is the local interest intent information, and the global decoding result output by the second global decoder is the global interest intent information.

[0120] Further, in the embodiment, the prediction process is to predict the dynamic interest intent information of the user according to the local interest intent information of the target geographic object output by the multiple second decomposition decoders and / or the global interest intent information of the target geographic object output by the second global decoder.

[0121] In the embodiment, the dynamic interest intent information of the user can be predicted in different ways. The following gives several implementation manners, but is not limited thereto.

[0122] In an embodiment, the dynamic interest intention information is generated by relying on a second global decoder. In this embodiment, each second encoder encodes a second data vector of a different information modality, and the second data vectors are fused and globally decoded by the second global decoder to generate the final dynamic interest intention information. The second global decoder integrates data of each information modality and outputs a final dynamic interest intention information by learning the relationship between the information modalities.

[0123] In another embodiment, the final dynamic interest intention information is generated by relying on local decoding results of second decomoders. Each second decomoder locally decodes the second data vector corresponding thereto to generate a local decoding result corresponding to each information modality. The output of each second decomoder represents the independent contribution of the information modality to the final dynamic interest intention information. Then, the local decoding results are fused by using weighted summation to obtain the final dynamic interest intention information. The final dynamic interest intention information is the weighted fusion result of the outputs of the second decomoders.

[0124] In yet another optional embodiment, the final dynamic interest intention information is generated by combining the second decomoders and the second global decoder. The second data vectors corresponding to the information modalities are locally decoded by the second decomoders corresponding thereto, and are finally fused and globally decoded in the second global decoder to generate the final dynamic interest intention information.

[0125] Further, in the embodiments of the present application, candidate information objects in a target application scenario are combined to jointly recommend information to the user according to the steady demand potential information and the dynamic interest intention information of the user. Optionally, the steady demand potential information and the dynamic interest intention information of the user are input into an information recommendation model, and the following operations are performed in the information recommendation model: the steady demand potential information and the dynamic interest intention information of the user are weighted and fused by using a first attention network in the information recommendation model to generate a joint embedding vector representing the current information demand of the user; attention calculation is performed on the joint embedding vector and information vectors corresponding to the candidate information objects by using a second attention network in the information recommendation model to obtain matching degrees of the joint embedding vector and each candidate information object; and a target information is selected from the candidate information based on the matching degrees by using a recommendation module in the information recommendation model, and the target information is output.

[0126] By inputting the steady-state demand potential information and the dynamic interest intention information of the user into the information recommendation model, and using the first attention network to weight and fuse the two types of demand information, a joint embedding vector that can better reflect the current information demand of the user can be generated. Further, the second attention network calculates the matching degree according to the joint embedding vector and the information vector of the candidate information object, so that the recommendation process can adaptively focus on the different features of different candidate information objects. Therefore, the embodiment simultaneously considers the long-term demand and short-term interest changes of the user in the selection of the candidate information object, and significantly improves the accuracy and pertinence of the recommendation result, In the embodiment, the first attention network in the information recommendation model is used to weight and fuse the user demand information of different time scales. Specifically, the first attention network weights and fuses the two types of data of different time scales according to the steady-state demand potential information and the dynamic interest intention information, to obtain a joint embedding vector that can comprehensively represent the user demand. The joint embedding vector comprehensively represents the demand performance of the user at the long-term and short-term time scales, and realizes the unified representation of the current information demand of the user.

[0127] In an optional embodiment, when the first attention network is used to weight and fuse the steady-state demand potential information and the dynamic interest intention information of the user to generate a joint embedding vector representing the current information demand of the user, the following steps are included: the steady-state demand potential information and the dynamic interest intention information of the user are input into the first attention network, and the following operations are performed in the first attention network: a gating signal is generated according to the dynamic interest intention information, and the steady-state demand potential information is element-wise modulated using the gating signal to obtain an activated steady-state demand potential vector; cross-attention calculation is performed on the activated steady-state demand potential vector and the dynamic interest intention information to obtain a context semantic vector; the activated steady-state demand potential vector, the context semantic vector, and the global aggregation representation of the dynamic interest intention information are spliced and then input into multiple parallel fusion sub-networks to obtain multiple candidate fusion vectors; routing weights are generated based on the steady-state demand potential information and the dynamic interest intention information, and the multiple candidate fusion vectors are weighted and summed according to the routing weights to obtain the joint embedding vector.

[0128] Thus, the first attention network modulates the steady-state demand potential information element by element through the gating signal, so that the features with higher relevance to the dynamic interest intention information are highlighted; then, the context semantic vector is obtained through cross-attention calculation to express the association between the two types of demand information; the activated steady-state demand potential vector, the context semantic vector and the global aggregation representation of the dynamic interest intention information are input into multiple parallel fusion sub-networks to generate candidate fusion vectors; and the routing weights generated based on the two types of demand information are used to weight and sum the candidate fusion vectors. The joint embedding vector obtained through the above processing can more accurately represent the current information demand of the user and improve the accuracy of subsequent recommendation matching.

[0129] In the present embodiment, the gating signal is generated according to the dynamic interest intention information, and the steady-state demand potential information is modulated element by element using the gating signal to obtain the activated steady-state demand potential vector. The gating signal is used to adjust the contribution of each component of the steady-state demand potential information in the feature dimension, so that the steady-state demand features with higher relevance to the current dynamic interest intention information are amplified in the element-by-element modulation process, and the irrelevant or weakly relevant features are relatively suppressed, so that the activated steady-state demand potential vector represents the steady-state demand expression after reweighting in the current dynamic interest context, providing a more demand state consistent steady-state demand input for subsequent cross-attention calculation with dynamic interest intention information.

[0130] In the present embodiment, in the case where the activated steady-state demand potential vector is obtained, the first attention network can further perform cross-attention calculation on the activated steady-state demand potential vector and the dynamic interest intention information to obtain the context semantic vector. The cross-attention calculation can be understood as an attention mechanism for modeling the association between different source features. The context semantic vector obtained through the cross-attention mechanism can be regarded as a semantic association enhanced representation of the activated steady-state demand potential vector in the dynamic interest context.

[0131] Further, the first attention network can concatenate the activated steady-state demand potential vector, the above-mentioned context semantic vector and the global aggregated representation of the dynamic interest intention information, and input the concatenated feature vector into a plurality of parallel fusion sub-networks. The plurality of fusion sub-networks generate corresponding candidate fusion vectors. In the case of obtaining a plurality of candidate fusion vectors, the corresponding routing weights are generated based on the steady-state demand potential information and the dynamic interest intention information, and the routing weights are used to indicate the contribution proportion of each candidate fusion vector in the final fusion result. The routing weight can be understood as a set of adaptive weighting coefficients derived from the joint features of the steady-state demand potential information and the dynamic interest intention information. By performing weighted summation on each candidate fusion vector, the final joint embedding vector can be obtained. The joint embedding vector simultaneously fuses long-term steady-state demand, short-term dynamic interest and semantic association information between the two, and can be used as a unified representation of the current information demand state of the user, providing direct input for subsequent candidate information object matching and joint recommendation.

[0132] Further, in the embodiments of the present application, the second attention network in the information recommendation model is used to perform attention calculation on the joint embedding vector and the information vector corresponding to the candidate information object to obtain the matching degree of the joint embedding vector and each candidate information object. In an optional embodiment, the joint embedding vector is input into the second attention network, and the following operations are performed in the second attention network: performing semantic analysis on the information vector of each candidate information object to obtain the sub-vector representation of each candidate information object at at least two different semantic granularity levels; calculating the absolute matching degree of each candidate information object and the joint embedding vector according to the joint embedding vector and the sub-vector representation of each candidate information object at each semantic granularity level; constructing the competitive context representation of each candidate information object relative to other candidate information objects based on the information vector of each candidate information object; performing matching degree calculation on the joint embedding vector and the competitive context representation of each candidate information object to obtain the relative matching degree of each candidate information object and the joint embedding vector considering the competitive relationship; and generating the matching degree of each candidate information object and the joint embedding vector according to the absolute matching degree and the relative matching degree of each candidate information object and the joint embedding vector.

[0133] By inputting the joint embedding vector into the second attention network and performing semantic analysis on the information vector of the candidate information object, a sub-vector representation of each candidate information object at different semantic granularity levels can be formed, and the hierarchy of semantic expression can be improved. The absolute matching degree calculated based on the joint embedding vector and the multi-granularity sub-vector can reflect the direct matching degree between the candidate information object and the current information demand of the user. Meanwhile, the competitive context representation constructed using the information vector of the candidate information object and the relative matching degree calculated therefrom can reflect the relative competitive position of the candidate information object within the candidate set. Furthermore, the matching degree generated by combining the absolute matching degree and the relative matching degree can select the target information while taking into account the direct matching degree and the relative competitive position within the set, thereby greatly improving the recommendation effect.

[0134] In this optional embodiment, the second attention network can perform semantic analysis on the information vector corresponding to each candidate information object to obtain a sub-vector representation of the candidate information object at at least two different semantic granularity levels. The semantic granularity level can be understood as different levels of abstraction or different content dimensions of the information vector of the candidate information object. The semantic granularity levels can be constructed as a hierarchical structure from coarse to fine, specifically including a coarse-grained level, a medium-grained level, and a fine-grained level, which are respectively used to represent the category-level semantics, the brand or attribute-level semantics, and the specific instance-level semantics of the candidate information object.

[0135] In this embodiment, according to the joint embedding vector and the sub-vector representation of each candidate information object at each semantic granularity level, the absolute matching degree of each candidate information object and the joint embedding vector is calculated. The absolute matching degree can be used to represent the direct matching degree of a single candidate information object itself and the joint embedding vector.

[0136] In an optional embodiment, in the calculation of the absolute matching degree of each candidate information object and the joint embedding vector according to the joint embedding vector and the sub-vector representation of each candidate information object at each semantic granularity level, the following steps are included: for each candidate information object, the joint embedding vector is matched with the sub-vector representation of the candidate information object at each semantic granularity level to obtain the local matching degree of the candidate information object at each semantic granularity level and the joint embedding vector; based on the joint embedding vector and / or the context information of the candidate information object, a fusion weight of the candidate information object at each semantic granularity level is dynamically generated through a gating network; and the local matching degrees of the candidate information object at each semantic granularity level and the joint embedding vector are weighted and fused according to the fusion weight to obtain the absolute matching degree of the candidate information object and the joint embedding vector.

[0137] By performing matching degree calculation between the joint embedding vector and the sub-vector representation of the candidate information object at different semantic granularity levels respectively, multiple local matching degrees of the candidate information object at fine-grained levels can be obtained, thereby reflecting the relevance of the candidate information object at different semantic granularities respectively. Further, based on the fusion weight generated based on the joint embedding vector and / or the context information of the candidate information object, the contribution proportion of each semantic granularity level in the final matching calculation can be adaptively adjusted, thereby highlighting the more representative semantic level. Finally, the absolute matching degree obtained by weighting and fusing the local matching degrees at each level can more comprehensively and accurately represent the direct matching degree between the candidate information object and the current information demand of the user. Thus, the absolute matching degree can more accurately distinguish the relevance of different candidate information objects in the candidate set, avoiding the traditional dependence on a single semantic level and improving the discrimination ability and accuracy of the recommendation result.

[0138] In the embodiment, for each candidate information object, the second attention network can perform matching degree calculation between the joint embedding vector and the sub-vector representation of the candidate information object at each semantic granularity level, thereby obtaining the local matching degree of the candidate information object at each semantic granularity level with the joint embedding vector. The local matching degree is used to represent the semantic matching degree between the semantic sub-representation of the candidate information object and the current information demand of the user (represented by the joint embedding vector) at the corresponding semantic granularity level.

[0139] In the embodiment, the second attention network can dynamically generate the fusion weight of the candidate information object at each semantic granularity level based on the joint embedding vector and / or the context information of the candidate information object through a gating network. The fusion weight is used to adjust the contribution degree of different semantic granularity levels in the calculation of the absolute matching degree, and can be implemented as a set of adaptive level weighting coefficients.

[0140] With the obtained fusion weight, the second attention network can weight and fuse the local matching degrees of the candidate information object at each semantic granularity level with the joint embedding vector according to the fusion weight, thereby obtaining the absolute matching degree of the candidate information object with the joint embedding vector. By weighting the local matching degrees of each semantic granularity level using the fusion weight, the absolute matching degree can comprehensively reflect the matching information of the candidate information object at different semantic granularity levels, and adaptively adjust the influence degree of each semantic level on the final fusion result, so that the absolute matching degree can accurately depict the direct matching degree between the candidate information object and the joint embedding vector.

[0141] Further, in order to depict the relative competition relationship of the candidate information objects within the candidate set, the second attention network can further construct a competition context representation of each candidate information object relative to other candidate information objects based on the information vector of the candidate information object. The competition context representation is a differentiated representation obtained by comparing and modeling the current candidate information object based on the features of other candidate information objects, and is used to represent the relative competition status.

[0142] In the embodiment, after obtaining the competition context representation of each candidate information object, the second attention network can perform matching degree calculation on the joint embedding vector and the competition context representation to obtain the relative matching degree of each candidate information object and the joint embedding vector. The relative matching degree is used to measure the degree of fit between the current candidate information object and the user's current information demand under the premise of other candidate information objects, thereby reflecting the relative priority of the candidate information object within the candidate set.

[0143] Further, according to the absolute matching degree and the relative matching degree of each candidate information object and the joint embedding vector, a matching degree of each candidate information object and the joint embedding vector is generated. The matching degree comprehensively considers the direct matching degree between the candidate information object and the joint embedding vector and the relative competition status of the candidate information object in the candidate set, and is used to comprehensively represent the overall matching situation of each candidate information object and the joint embedding vector.

[0144] In an optional embodiment, based on the geographic semantic features, distance features and geographic context features of the geographic object description information, the geographic object description information in the ontology library is data enhanced by using the geographic object description information in the third-party library, including: S11, for any geographic object description information in the third-party library, recall of similar geographic object description information in the ontology library is performed to obtain at least two geographic object description information, the geographic object description information at least including name information, address information and spatial position information of the geographic object; S12, using a geographic semantic feature encoder, the geographic semantic similarity between the at least two geographic object description information is calculated based on the name information and the address information in the at least two geographic object description information; S13, using a distance feature encoder, the spatial distance features between the at least two geographic object description information are calculated based on the spatial position information in the at least two geographic object description information; S14, using a geographic context feature encoder, the geographic context similarity between the at least two geographic object description information is calculated based on the at least two geographic object description information and the respective surrounding geographic environment information; S15, input the geographic semantic similarity, the spatial distance feature and the geographic context similarity into a classification head network, and output a classification result of whether the at least two geographic object description information are directed to a same geographic object; S16, according to the classification result, perform data enhancement on the geographic object description information in the ontology library by using the geographic object description information belonging to the third-party library in the at least two geographic object description information, to obtain the target library.

[0145] In the embodiment, for each geographic object description information in the third-party library, at least one similar geographic object description information in the ontology library is recalled based on a rule matching strategy and / or a text similarity matching strategy, to form the at least two geographic object description information.

[0146] In some embodiments, similar geographic object description information is screened (or recalled) in some ways as the at least two geographic object description information to be processed in the embodiment. For example, any geographic object description information in the ontology library can be taken as a reference, and similar geographic object description information is recalled from each third-party library, and then the any geographic object description information and all the similar geographic object description information recalled are taken together as the at least two geographic object description information to be processed in the embodiment. For another example, any geographic object description information in the ontology library can be taken as a reference, and similar geographic object description information is recalled from each third-party library, and then for each third-party library, the any geographic object description information and the similar geographic object description information recalled from the third-party library are taken together as the at least two geographic object description information to be processed in the embodiment. For another example, for any third-party library, any geographic object description information in the third-party library can be taken as a reference, and similar geographic object description information is recalled from the ontology library, and then the any geographic object description information and all the similar geographic object description information recalled from the ontology library are taken together as the at least two geographic object description information to be processed in the embodiment. For another example, for any third-party library, any geographic object description information in the third-party library can be taken as a reference, and similar geographic object description information is recalled from the ontology library, and then the any geographic object description information and one similar geographic object description information recalled from the ontology library are taken together as the at least two geographic object description information to be processed in the embodiment.

[0147] In this embodiment, the geographic object description information includes name information, address information, and spatial location information. To fully utilize the geographic object description information, in this embodiment, a neural network model is used to extract features from the name and address semantics, spatial location information, and surrounding geographic environment, respectively, to fully utilize the features extracted from different information dimensions, and the neural network model is used to accurately identify the geographic object based on the extracted features.

[0148] It should be noted that the implementation form of the neural network model used in the embodiments of the present application is not limited, and can be various neural network models based on deep learning. Alternatively, each neural network model used can be a deep learning model with relatively small model parameter size, or a deep learning model with relatively large model parameter size can be used, wherein the large model is only an example of a model, and the embodiments of the present application do not limit the number of model parameters supported by the deep learning model used to meet the actual needs. The deep learning model involved in the embodiments of the present application is not limited to a specific type, and different models can be selected according to the implementation needs. For example, a language model (Language Model, LM) based on artificial intelligence can be used, and of course a multimodal model (Multimodal Model, MM) capable of processing multiple modal information can also be used.

[0149] In this embodiment, the neural network model used at least includes a geographic semantic feature encoder, a distance feature encoder, a geographic context feature encoder, and a classification head network. These neural network models cooperate with each other and work together to complete the judgment of whether the description information of at least two geographic objects points to the same geographic object, and each model contributes to improving the judgment accuracy. The role of each model is described in detail below.

[0150] In the embodiments of the present application, on the one hand, the geographic semantic feature encoder is used to calculate the geographic semantic similarity between the at least two geographic object description information based on the name information and the address information in the at least two geographic object description information.

[0151] The name information and the address information in the geographic object description information are relatively text type information. Therefore, the name information and the address information in the at least two geographic object description information are input to the geographic semantic feature encoder to calculate the semantic similarity between the at least two geographic object description information using the geographic semantic feature encoder.

[0152] For example, taking the example of a cell as a geographic object, geographic object description information corresponding to two cells is obtained from different database sources. For the geographic object description information of each cell, the corresponding name information and address information are extracted. In some embodiments, for the extracted name information and address information, a structured input sequence containing preset markers is formatted, for example, [CLS] name: <name information> address: <address information> [SEP]. Wherein, [CLS] is used to represent the beginning of the sequence, and can also be regarded as a classification marker, representing the aggregation information of the entire sequence. [SEP] is used to separate different geographic object description information, which can be called a segmentation marker. <name information> and <address information> refer to the name information and address information of the geographic object, respectively.

[0153] In this embodiment, geographic object description information A1 is referred to as A1, the corresponding name information is A cell, and the corresponding address information is XX Road XX. Geographic object description information A2 is referred to as A2, the corresponding name information is A cell 2, and the corresponding address information is YY Road YY. Optionally, any of the above geographic object description information is formatted as: [CLS] name: <name information> address: <address information> [SEP]. For example, [CLS] name: A cell; address: XX Road XX [SEP]. Wherein, [CLS] and [SEP] are two markers, [CLS] is used to represent the beginning of the input sequence, and the corresponding final hidden state is used as the global semantic representation vector of the entire input sequence pair, which facilitates the geographic semantic feature encoder to output the global semantic representation vector for subsequent similarity calculation or classification decision; [SEP] is used to separate the description information of different geographic objects in the input sequence, so that the geographic semantic feature encoder can distinguish the text boundaries of different geographic object description information, and effectively model the semantic similarity between different geographic object description information.

[0154] In this embodiment, the geographical semantic feature encoder is used to semantically encode at least two geographical object description information, and geographical semantic similarity between the at least two geographical object description information is obtained. For example, the geographical semantic feature encoder can extract a global semantic representation vector at the [CLS] position, which is used to calculate the geographical semantic similarity between the at least two geographical object description information. For another example, the semantic representation vectors at all positions can also be taken and weightedly fused to calculate the geographical semantic similarity between the at least two geographical object description information. The geographical semantic similarity between the at least two geographical object description information deeply fuses the name information and the address information of the at least two geographical object description information, and captures the semantic similarity therebetween. In this embodiment, on one hand, the distance feature encoder is used to calculate the spatial distance feature between the at least two geographical object description information based on the spatial position information in the at least two geographical object description information. In this embodiment, the spatial position information in the at least two geographical object description information is taken as the input of the distance feature encoder, and the distance feature encoder is used to calculate the spatial distance feature between the at least two geographical object description information.

[0155] In this embodiment, the spatial position information refers to information describing the spatial position of a geographical object, which can be implemented as the latitude and longitude information of the geographical object, but is not limited thereto. In addition to the latitude and longitude information, the spatial position information can also be implemented as relative position information reflecting the spatial relationship between the geographical object and the surrounding roads, landmarks or other reference objects.

[0156] In an optional embodiment, taking the latitude and longitude of the geographical object as the spatial position information, the Haversine distance or the spherical cosine distance between the latitude and longitude of any two geographical objects can be calculated, and the spatial distance feature between any two geographical objects can be determined based on the calculation result.

[0157] In addition, if two address entity description information point to the same geographical object in the real world, their surrounding geographical environment information will be highly coincident. The surrounding geographical environment information can be, for example, POI (Point of Interest), such as a store, a school, a bus stop, and the like, or road structure information such as road density, road type, traffic hub position, and the like, without limitation.

[0158] In this embodiment, according to the address information in each geographical object description information, a map service interface is called to obtain the surrounding environment information within a preset distance range around the geographical object described by the geographical object description information. The preset distance range can be 300 meters, 500 meters, 1 kilometer, and the like, without limitation, and can be flexibly set according to the accuracy requirement or the coverage requirement, and the like.

[0159] Further, the geographical context feature encoder is used to calculate geographical context similarity between the at least two geographical object description information based on the at least two geographical object description information and the respective surrounding geographical environment information.

[0160] Specifically, by using the geographical context feature encoder, not only the description information of each geographical object and the corresponding geographical environment information can be fused to obtain a feature vector representing the context of the geographical object, but also the feature interaction processing can be performed on the at least two geographical context feature vectors to construct a feature representation reflecting the context relationship between the at least two geographical objects.

[0161] The feature interaction processing can include, but is not limited to, concatenating, subtracting, or combining the concatenation and subtraction of the at least two geographical context feature vectors. Through the feature interaction processing, a feature vector representing the geographical context similarity between the at least two geographical objects described by the geographical object description information is generated.

[0162] In this embodiment, the geographical semantic similarity, the spatial distance feature, and the geographical context similarity are input into the classification head network, and a classification processing is performed in the classification head network to finally output a classification result of whether the at least two geographical object description information points to the same geographical object.

[0163] In an optional embodiment, the classification head network is implemented as a total classification head network, and the geographical semantic similarity feature, the spatial distance feature, and the geographical context similarity feature can be concatenated to form a fusion feature vector, which is input into the classification head network for classification to output a classification result of whether the two geographical object description information points to the same geographical object. In this optional embodiment, the classification head network can adopt a multi-layer perceptron structure. The total classification head network has the advantages of simple implementation and high efficiency.

[0164] In another optional embodiment, a hierarchical fusion architecture classification head network is designed, i.e., the classification head network includes multiple sub-classification head networks and a total classification head network. The geographical semantic similarity feature, the spatial distance feature and the geographical context similarity feature can be respectively input into different sub-classification head networks to obtain corresponding initial classification results. Further, the output results of the multiple sub-classification head networks are fused and predicted by the total classification head network to obtain a classification result, which is used to represent whether the two geographical object description information point to the same geographical object. Further, optionally, in the process of fusion and prediction of the output results of the multiple sub-classification head networks by the total classification network, the influence of the geographical semantic similarity feature, the spatial distance feature and the geographical context similarity feature can also be considered, i.e., the geographical semantic similarity feature, the spatial distance feature and the geographical context similarity feature, and the initial classification results output by the multiple sub-classification head networks are input into the total classification head network for fusion and prediction to obtain a final classification result. The multiple sub-classification head networks generate local discriminant results based on different semantic dimensions of the geographical object description information, and the total classification head network performs nonlinear fusion and confidence calibration on the local discriminant results to output a global consistency classification decision, which can effectively overcome the defects of single semantic dimension being easily affected by noise interference or information loss, and significantly improve the accuracy, robustness and scene adaptability of geographical object matching.

[0165] In this embodiment, according to the classification result, the geographical object description information in the geographical object description information ontology library belonging to the third-party library among the at least two geographical object description information is used for data enhancement to obtain a target library.

[0166] In this embodiment, if the classification result indicates that the at least two geographical object description information point to the same geographical object, the geographical object description information belonging to the third-party library and the geographical object description information belonging to the ontology library among the at least two geographical object description information are associated to obtain a target library. Specifically, the geographical object description information of the third-party library is taken as supplementary geographical object description information of the ontology library geographical object of the same geographical object, so as to associate the geographical description information of the third-party library with the geographical description information of the ontology library to form a target library. After the association processing, the generated target library will contain geographical object description information from the third-party library and the ontology library. In this way, the mounting rate and accuracy of the target library are improved. If the at least two geographical object description information point to different geographical objects, the geographical object description information in the third-party library is added to the ontology library. In this way, the geographical object description information of the ontology library can be expanded.

[0167] In the embodiment of the present application, for at least two geographic object description information, a geographic semantic feature encoder, a distance feature encoder and a geographic context feature encoder are used respectively to encode the name and address semantics, spatial position and surrounding geographic environment to obtain geographic semantic similarity, spatial distance feature and geographic context similarity; the geographic semantic similarity, spatial distance feature and geographic context similarity are input into a classification head network to output a classification result of whether the at least two geographic object description information points to the same geographic object; and according to the classification result, data augmentation is performed on the geographic description information of the ontology library to obtain a target library. In this way, by introducing the learning of the correlation between different features by the model, the multi-dimensional features such as geographic semantics, spatial position and surrounding geographic environment are fully utilized, and feature fusion and accurate judgment are realized in the classification head network, thereby improving the accuracy and robustness of geographic object matching; and the mounting rate and accuracy of the target library are improved.

[0168] In the embodiments of the present application, a model training method of a first or second information prediction model is provided. The first or second information prediction model can be obtained by training based on the model training method. The training process of the first information prediction model is taken as an example for description. The training method of the second information prediction model can be trained by referring to the following model training method, and the difference lies in that the training sample data and the target result are different.

[0169] A model training method includes the following steps: S21, obtaining training sample data, the training sample data corresponding to a target result in a target application scenario, the target result being preset value information corresponding to the training sample data; S22, classifying the training sample data according to the information modalities presented by the training sample data; S23, inputting the training sample data into an initial model, the initial model including a first global decoder, a plurality of first encoders corresponding to a plurality of information modalities respectively, and a first decomposition decoder, and the training process including: S24, encoding process: inputting the training sample data into the plurality of first encoders corresponding to the plurality of information modalities respectively according to the above classification categories to generate a plurality of data vectors corresponding to the plurality of information modalities respectively; S25, decoding process: inputting the plurality of data vectors corresponding to the plurality of information modalities respectively into the corresponding first decomposition decoders for local decoding, and inputting the plurality of data vectors corresponding to the plurality of information modalities after fusion into the first global decoder for global decoding; S26, a tuning process: adjusting the parameters of the corresponding first encoder according to the sub-residuals between the local decoding results of each of the plurality of first decoders and the target result, and adjusting the parameters of the plurality of first encoders according to the total residual between the first global decoding result and the target result; iteratively performing the encoding process and the decoding process until the total residual meets a set value to obtain the first information prediction model.

[0170] In the embodiment, the application scenarios of the above model training method are not limited. For example, it can be a scenario of generating images and texts, information prediction, information recommendation, e-commerce marketing, auction marketing, and various information evaluation. The target result refers to an expected result corresponding to the training sample data, which is a labeling result of the training sample data. The training sample data and the target result will be different according to different application scenarios and model functions. For example, in the information recommendation scenario, an information recommendation model can be trained according to the training sample data, and correspondingly, the target result can be a recommended result corresponding to the training sample data. Specifically, taking the recommendation of commodity information in the e-commerce marketing scenario as an example, the training sample data can be user portrait data, historical behavior data, etc., and the corresponding target result can be recommended commodity information. Therefore, the method of the embodiment of the application can train a commodity information recommendation model capable of recommending commodity information to users based on user portrait data, historical behavior data, etc.

[0171] In the embodiment, the training sample data is a data set for model training. In the embodiment, each training sample data includes multiple types of data, and the multiple types of data respectively present different information modalities. The information modality refers to the form of data. In some embodiments of the application, the information modality can be referred to as modality. The different types of data are classified to obtain information data of multiple modalities.

[0172] Different application scenarios have corresponding target results. In the image and text generation scenario, the training sample data includes two information modalities of image data and text prompt words, the model input is an image and a corresponding text description, and the output is a generated image consistent with the semantics of the prompt words. In the commodity recommendation scenario, the training sample data includes user attribute information, user historical behavior data, and commodity image and text information; the model input is user portrait, historical behavior, and commodity content features, and the target result is the labeling of the user's interaction result with the target commodity, such as click, collection, or purchase behavior indicators. In the information evaluation scenario (for example, geographic object value evaluation), the training sample data includes location, supporting environment, building features, etc.; the model input is a multi-modality feature set of a geographic object, and the target result is a labeled quantitative result, such as a numerical evaluation value.

[0173] Further, the training sample data is input into an initial model. The initial model includes a first global decoder, a plurality of first encoders corresponding to the plurality of information modalities, and a plurality of first decomoders corresponding to the plurality of information modalities. Each information modality has a dedicated first encoder and a dedicated first decomoder.

[0174] In this embodiment, the classified different modality data is respectively input into each corresponding first encoder, and through the encoding process, a data vector of each modality is generated.

[0175] In this embodiment, the data vector corresponding to each modality is respectively input into the corresponding first decomoder for local decoding. The first decomoder is responsible for generating a local decoding result according to the encoding vector of each modality. After the data vectors corresponding to the plurality of modalities are fused, a fusion vector is obtained. Further, the fusion vector is input into the global decoder for global decoding to obtain a global decoding result.

[0176] In this embodiment, global decoding and local decoding are relative concepts, and essentially belong to the results obtained after the model decodes the input features. The local decoding refers to the decoding process independently completed by each modality at the level of the first decomoder, which is used to restore or reconstruct the information features corresponding to the modality. The local decoding corresponds to a local decoding result. The global decoding refers to the decoding process uniformly performed by the first global decoder after the encoding results of the modalities are fused. The global decoding obtains a global decoding result.

[0177] In an example, in a text-to-image generation scenario, the local decoding result and the global decoding result can be expressed as image data.

[0178] The first decomoders corresponding to the information modalities can respectively perform local decoding on the structural information, texture details, or semantic layout of the image to form a plurality of local decoding results. The first global decoder generates the fusion vector based on the fused fusion vector and outputs a generated image consistent with the semantic of the prompt.

[0179] In this embodiment, the parameters of the corresponding first encoders are adjusted according to the sub-residuals between the local decoding results of the plurality of first decomoders and the target results. The sub-residual refers to the error between the local decoding result of the first decomoder and the target result. The sub-residual is fed back to the corresponding first encoder through back propagation to adjust the parameters of the first encoder to reduce the error.

[0180] In this embodiment, the parameters of the plurality of first encoders are adjusted according to the total residual between the global decoding result and the target result. The total residual is the error between the global decoding result and the target result, and the parameters of the plurality of first encoders are adjusted after back propagation.

[0181] In the embodiment, the manner of adjusting the first encoder parameters according to the total residual error and the sub residual error is not limited.

[0182] In an optional embodiment, the corresponding first encoders are adjusted by using the sub residual error first, and then optimized by using the total residual error. The sub residual error can quickly correct the feature learning of a single modality at a local level, so that the encoding representation of each information modality is more accurate and stable. After the sub residual error adjustment is completed, the total residual error between the global decoding result and the target result is used to uniformly adjust all the first encoders, so as to coordinate the feature distribution between the modalities from a global level. Through the local-global optimization sequence, the model can be optimized from "intra-modality refinement" to "inter-modality fusion" layer by layer, and the overall training efficiency and the final output consistency are improved.

[0183] In yet another optional embodiment, the sub residual error corresponding to the plurality of first decoders and the total residual error corresponding to the first global decoder can be jointly calculated to generate a comprehensive target residual error. The target residual error can fuse the sub residual error and the total residual error according to a preset weight coefficient. The model adjusts the parameters of each first encoder according to the target residual error, so as to consider the accuracy of the local modality feature and the consistency of the global feature expression, and improve the generalization performance of the model in the multi-modal task.

[0184] In yet another optional embodiment, the adjustment sequence can be dynamically determined according to the relative size of the sub residual error and the total residual error in the training process. For example, when the sub error of a single modality is large, the sub residual error is used to adjust the parameters of the corresponding first encoder first; when the total residual error is large, the total residual error is used for overall optimization. By evaluating the change trend of the sub residual error and the total residual error, the optimization sequence is adaptively determined, so as to improve the effectiveness of the parameter adjustment and the training convergence speed.

[0185] In the embodiment, the encoding process and the decoding process are iteratively executed. The total residual error after each training is calculated and compared with a preset setting value, until the total residual error meets the setting value, and then the model training can be ended to obtain a target model.

[0186] In this application embodiment, a hybrid model architecture of multiple encoders, multiple sub-decoders, and a global decoder is proposed. Sample data of different information modalities correspond to different encoders and sub-decoders, transforming the mode of a single model processing all training sample data into a divide-and-conquer mode. The inherent complexity of each encoder and sub-decoder is low, which can reduce the complexity of model training and save resource consumption. Moreover, different encoders and sub-decoders can be trained in parallel, which is beneficial to shorten the model training time. Furthermore, by utilizing the dual decoding mechanism of sub-decoders and global decoders, the encoder parameters can be continuously adjusted through local and global optimization to optimize the model performance, improve the accuracy of inference results, and accelerate the model convergence speed, thereby further improving the model training efficiency and saving model training time.

[0187] It should be noted that the execution subject of each step of the method provided in the above embodiments can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 701 to 703 can be device A; or the execution subject of steps 701 and 702 can be device A, and the execution subject of step 703 can be device B; and so on.

[0188] Furthermore, in some of the processes described in the above embodiments and accompanying drawings, multiple operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or they may be executed in parallel. The operation numbers, such as 701, 702, etc., are merely used to distinguish different operations and do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0189] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 9 As shown, in practice, this electronic device includes a memory 94 and a processor 95.

[0190] Memory 94 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0191] The processor 95 is coupled with the memory 94 and used for executing a computer program in the memory 94, so as to acquire portrait data and historical behavior data of a user, the portrait data comprising a target geographical object associated with the user, attribute data of the target geographical object reflecting preliminary value information of the target geographical object, and the historical behavior data reflecting a dynamic interest intention of the user; determine a similar geographical object of the target geographical object and a quantity of the similar geographical object according to the attribute data of the target geographical object, the attribute data of the target geographical object at least comprising a target area to which the target geographical object belongs; generate a value trend change coefficient of the target area according to a fitting curve of historical transaction prices and time relationship of other geographical objects in the target area; determine a missing degree of the similar geographical object according to the quantity of the similar geographical object and a quantity of transaction objects in the target area; adjust the preliminary value information by taking the missing degree as an adjustment coefficient and a product of a statistical value of historical transaction prices of the similar geographical object and the trend change coefficient as an adjustment amount, so as to obtain target value information reflecting a steady-state demand potential of the user; and perform joint information recommendation to the user from two dimensions of the dynamic interest intention and the steady-state demand potential according to the historical behavior data and the target value information of the target geographical object.

[0192] The detailed implementation of each operation described above can be referred to the corresponding description in the foregoing embodiments, which will not be repeated here.

[0193] Further, as shown in Figure 9 , the electronic device further includes a communication component 96, a display 97, a power supply component 98, an audio component 99, and other components. Figure 9 Some components are only schematically shown in the electronic device, and it does not mean that the electronic device only includes Figure 9 the components shown. In addition, Figure 9 the components in the dashed box are optional components, not mandatory components, and the specific implementation can be determined according to the product form of the electronic device. The electronic device of the embodiment can be implemented as a terminal device such as a desktop computer, a notebook computer, a smart phone or an IOT device, or a server device such as a general server, a cloud server or a server array. If the electronic device of the embodiment is implemented as a terminal device such as a desktop computer, a notebook computer or a smart phone, it can include Figure 9 the components in the dashed box; if the electronic device of the embodiment is implemented as a server device such as a general server, a cloud server or a server array, it can not include Figure 9 the components in the dashed box.

[0194] The above-described memory can be implemented by any type of volatile or nonvolatile memory devices or a combination thereof, such as a Static Random-Access Memory (SRAM), an Electrically Erasable Programmable Read Only Memory (EEPROM), an Erasable Programmable Read Only Memory (EPROM), a Programmable Read-Only Memory (PROM), a Read-Only Memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or a compact disk.

[0195] The above-described communication component is configured to facilitate wired or wireless communication between the device in which the communication component is located and other devices. The device in which the communication component is located can access a wireless network based on a communication standard, such as a 2G, 3G, 4G / LTE, 5G, or the like mobile communication network, or a combination thereof. In an example embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel.

[0196] The above-described display includes a screen, which can include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes the touch panel, the screen can be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense a touch, a slide, and a gesture on the touch panel. The touch sensor can not only sense a boundary of a touching or a sliding action, but also detect a duration and a pressure related to the touching or the sliding action.

[0197] The above-described power component provides power to various components of the device in which the power component is located. The power component can include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which the power component is located.

[0198] The above-described audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) that is configured to receive external audio signals when the device in which the audio component is located is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals can be further stored in the memory or transmitted via the communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0199] Accordingly, the embodiments of the present application also provide a computer readable storage medium storing a computer program, when the computer program is executed by a processor, the processor is enabled to implement each step in the above method embodiments. The computer readable storage medium includes volatile or non-volatile or their combination, and can be removable or non-removable. Examples of the computer readable storage medium include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital video disc (DVD) or other optical storage, magnetic cassette, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium Accordingly, the embodiments of the present application also provide a computer program product, the computer program product includes computer programs or instructions, when the computer programs or instructions are executed by a processor, the processor is enabled to implement each step in the above method embodiments. It should be understood that each process or combination of multiple processes in the above method flow can be implemented by the computer programs or instructions. In addition, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device, so that the processor of the general-purpose computer, the special-purpose computer, the embedded processor or other programmable data processing device can be implemented as a device for implementing the corresponding functions in the above method embodiments.

[0200] It should also be noted that the terms "comprising", "including", or any other variant thereof are intended to cover non-exclusive inclusion, so that processes, methods, articles or devices including a series of elements not only include those elements, but also include other elements not explicitly listed, or further include elements inherent in such processes, methods, articles or devices. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0201] The above merely provides an example of the present application, and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall fall into the scope of claims of the present application.

Claims

1. An information recommendation method, characterized in that, include: Acquire user profile data and historical behavior data, wherein the profile data includes the identification information of target geographic objects associated with the user; According to the identification information, the attribute data of the target geographic object is obtained from the target library. The target library is obtained by using the geographic semantic features, distance features and geographic context features of the geographic object description information to perform data augmentation on the geographic object description information in the ontology library using the geographic object description information in the third-party library. The geographic object description information includes the attribute data of the corresponding geographic object. Based on the attribute data of the target geographic object, predict the user's steady-state demand potential information, and based on the historical behavior data, predict the user's dynamic interest intention information; Based on the user's steady-state demand potential information and dynamic interest intent information, combined with candidate information objects in the target application scenario, joint information recommendation is performed for the user.

2. The method according to claim 1, characterized in that, Based on the attribute data of the target geographic object, predict the user's steady-state demand potential information, including: Based on the attribute data of the target geographic object, similar geographic objects are determined, wherein the attribute data of the target geographic object includes at least the target region to which the target geographic object belongs; Based on the attribute information of the target area, the attribute data of the target geographic object, and the attribute data of the similar geographic objects, a pre-trained first information prediction model is invoked to perform information prediction in order to obtain the steady-state demand potential information of the user.

3. The method according to claim 2, characterized in that, Based on the surrounding environmental information of the target area, the attribute data of the target geographic object, and the attribute data of similar geographic objects, a pre-trained first information prediction model is invoked to perform information prediction, thereby obtaining the user's steady-state demand potential information, including: The attribute information of the target area, the attribute data of the target geographic object, and the attribute data of the similar geographic objects are classified to obtain first data of multiple information modalities; The first data of the multiple information modalities are input into the first information prediction model, which includes a first global decoder, a first encoder and a first sub-decoder corresponding to the multiple information modalities, and the following operations are performed within the first information prediction model: Encoding process: The first data of the multiple information modalities are respectively input into the first encoder corresponding to the multiple information modalities for encoding, generating the first data vector corresponding to each of the multiple information modalities; Decoding process: The first data vectors corresponding to each of the multiple information modes are input to the corresponding first decoder for local decoding, and the first data vectors corresponding to each of the multiple information modes are fused and input to the first global decoder for global decoding; Prediction process: Based on the local value information of the target geographic object output by the plurality of first decoders and / or the global value information of the target geographic object output by the first global decoder, predict the steady-state demand potential information of the user.

4. The method according to claim 1, characterized in that, Based on the historical behavior data, predict the user's dynamic interest and intent information, including: Based on the information modalities presented by the historical behavior data, the historical behavior data is classified to obtain second data with multiple information modalities; The second data of the multiple information modalities are input into the second information prediction model, which includes a second global decoder, a second encoder and a second sub-decoder corresponding to the multiple information modalities, and the following operations are performed within the second information prediction model: Encoding process: The second data of the multiple information modalities are respectively input into the second encoder corresponding to the multiple information modalities for encoding, so as to generate the second data vector corresponding to each of the multiple information modalities; Decoding process: The second data vectors corresponding to each of the multiple information modes are input to the corresponding second sub-decoders for local decoding, and the second data vectors corresponding to each of the multiple information modes are fused and input to the second global decoder for global decoding; Prediction process: Based on the local interest intent information output by the multiple second sub-decoders and / or the global interest intent information output by the second global decoder, predict the user's dynamic interest intent information.

5. The method according to claim 3, characterized in that, Also includes: Acquire training sample data, wherein the training sample data corresponds to a target result in the target application scenario, and the target result is a preset value information corresponding to the training sample data; The training sample data is classified according to the information modality presented by the training sample data; The training sample data is input into an initial model, which includes a first global decoder, a first encoder and a first sub-decoder corresponding to multiple information modalities, and the training process includes: Encoding process: The training sample data are input into the first encoder corresponding to the multiple information modalities according to the above classification categories for encoding, so as to generate the data vectors corresponding to each of the multiple information modalities; Decoding process: The data vectors corresponding to each of the multiple information modes are input to the corresponding first decoder for local decoding, and the data vectors corresponding to each of the multiple information modes are fused and input to the first global decoder for global decoding; Optimization process: Adjust the parameters of the corresponding first encoder according to the sub-residual between the local decoding results of each of the multiple first decoders and the target result, and adjust the parameters of the multiple first encoders according to the total residual between the first global decoding result and the target result; The encoding and decoding processes are executed iteratively until the total residual satisfies a set value, so as to obtain the first information prediction model.

6. The method according to any one of claims 1-5, characterized in that, Based on the user's steady-state demand potential information and dynamic interest intent information, combined with candidate information objects in the target application scenario, joint information recommendation is performed for the user, including: The user's steady-state demand potential information and dynamic interest intention information are input into the information recommendation model, and the following operations are performed in the information recommendation model: Using the first attention network in the information recommendation model, the steady-state demand potential information and dynamic interest intention information of the user are weighted and fused to generate a joint embedding vector representing the user's current information demand; Using the second attention network in the information recommendation model, attention is calculated on the joint embedding vector and the information vector corresponding to the candidate information object to obtain the matching degree between the joint embedding vector and each candidate information object; Using the recommendation module in the information recommendation model, target information is selected from the candidate information based on the matching degree, and the target information is output.

7. The method according to claim 6, characterized in that, Using a first attention network, the user's steady-state demand potential information and dynamic interest intention information are weighted and fused to generate a joint embedding vector representing the user's current information demand, including: The user's steady-state demand potential information and dynamic interest intention information are input into the first attention network, and the following operations are performed in the first attention network: A gating signal is generated based on the dynamic interest intent information, and the gating signal is used to modulate the steady-state demand potential information element by element to obtain the activated steady-state demand potential vector. Cross-attention calculation is performed on the activated steady-state demand potential vector and the dynamic interest intent information to obtain the context semantic vector; The activated steady-state demand potential vector, the context semantic vector, and the global aggregated representation of the dynamic interest intent information are concatenated and then input into multiple parallel fusion sub-networks to obtain multiple candidate fusion vectors. Based on the steady-state demand potential information and the dynamic interest intent information, a routing weight is generated, and the multiple candidate fusion vectors are weighted and summed according to the routing weight to obtain the joint embedding vector.

8. The method according to claim 6, characterized in that, Using the second attention network in the information recommendation model, attention is calculated on the joint embedding vector and the information vectors corresponding to the candidate information objects to obtain the matching degree between the joint embedding vector and each candidate information object, including: The joint embedding vector is input into the second attention network, and the following operations are performed in the second attention network: Semantic parsing is performed on the information vector of each candidate information object to obtain the sub-vector representation of each candidate information object at at least two different semantic granularity levels; Based on the joint embedding vector and the sub-vector representation of each candidate information object at each semantic granularity level, calculate the absolute matching degree between each candidate information object and the joint embedding vector; Based on the information vector of each candidate information object, construct a competitive context representation of each candidate information object relative to other candidate information objects; The matching degree between the joint embedding vector and the competitive context representation of each candidate information object is calculated to obtain the relative matching degree between each candidate information object and the joint embedding vector under the consideration of the competitive relationship; The matching degree between each candidate information object and the joint embedding vector is generated based on the absolute and relative matching degree between each candidate information object and the joint embedding vector.

9. The method according to claim 8, characterized in that, Based on the joint embedding vector and the sub-vector representations of each candidate information object at various semantic granularity levels, the absolute matching degree between each candidate information object and the joint embedding vector is calculated, including: For each candidate information object, the matching degree of the joint embedding vector with the sub-vector representation of the candidate information object at each semantic granularity level is calculated to obtain the local matching degree of the candidate information object with the joint embedding vector at each semantic granularity level. Based on the joint embedding vector and / or the context information of the candidate information object, the fusion weights of the candidate information object at each semantic granularity level are dynamically generated through a gating network. The local matching degree between the candidate information object and the joint embedding vector at each semantic granularity level is weighted and fused according to the fusion weight to obtain the absolute matching degree between the candidate information object and the joint embedding vector.

10. The method according to any one of claims 1-5 or 7-8, characterized in that, Based on the geographic semantic features, distance features, and geographic context features of geographic object description information, data augmentation is performed on the geographic object description information in the ontology library using geographic object description information from third-party libraries, including: For any geographic object description information in the third-party library, similar geographic object description information is retrieved in the ontology library to obtain at least two geographic object description information, wherein the geographic object description information includes at least the name information, address information and spatial location information of the geographic object; Using a geographic semantic feature encoder, the geographic semantic similarity between the at least two geographic object descriptions is calculated based on the name and address information in the descriptions of the at least two geographic objects. Using a distance feature encoder, spatial distance features between the at least two geographic object descriptions are calculated based on spatial location information in the descriptions of the at least two geographic objects. Using a geographic context feature encoder, the geographic context similarity between the description information of the at least two geographic objects is calculated based on the description information of the at least two geographic objects and the geographic environment information of their respective surroundings. The geographic semantic similarity, spatial distance features, and geographic context similarity are input into the classification head network, and the classification result is output to determine whether the description information of the at least two geographic objects points to the same geographic object. Based on the classification results, the geographic object description information in the ontology library is augmented using the geographic object description information belonging to a third-party library from the at least two geographic object description information to obtain the target library.

11. An electronic device, characterized in that, include: Processor and memory; The memory stores a computer program, and after the processor runs the computer program, it performs the method described in any one of claims 1-10.

12. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, performs the method described in any one of claims 1-10.

13. A computer program product, characterized in that, include: Computer instructions, when executed by a processor, perform the method described in any one of claims 1-10.

Images