POI recommendation method, computer program product, storage medium and computer electronic device

Abstract

To provide a point of interest (POI) recommendation method based on a time-series knowledge map.SOLUTION: Constructing a dynamic time-series knowledge map and a static group knowledge map based on complete historical behavior trajectories of all users, and obtaining a historical behavior trajectory sub-string of the target user before the to-be-predicted moment, sequentially extracting user comment texts of each point of interest accessed by the user from the historical behavior trajectory sub-string, performing word embedding on the user comment texts by using an aspect-based emotion analysis module constructed based on a pre-trained model, and splicing emotion embeddings of all the user comment texts to obtain a user comment emotion embedding sequence, and inputting the historical behavior trajectory sub-string, the dynamic time-series knowledge map, the static group knowledge map, and the user comment emotion embedding sequence into a point of interest recommendation model to predict a point of interest accessible by the target user at a next moment.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to the field of personalized recommendation technology, and more particularly to a method for recommending points of interest (POIs) based on a time-series knowledge map. [Background technology]

[0002] The breakthrough development of the mobile internet has brought users around the world closer together, allowing them to share their daily activities and share the joys of life on geographically-based social platforms. The massive amount of user interaction information has also driven the emergence of geotagged datasets such as Foursquare, Gowalla, and Yelp, providing new vitality and impetus to point-of-interest (POI) recommendations. POI recommendations can use users' past check-in information to predict where they might next visit, while also leveraging multimodal information such as time, geographic location, POI category, and social connections to achieve better prediction capabilities and provide greater convenience for users' daily outings.

[0003] Among existing technologies, many map neural network-based methods have achieved good results by learning global user and POI features. However, most existing methods only target one of the elements such as trajectory information, geographic location, social network, and user comments, and do not fully utilize the advantages of multimodal information in real situations. In addition, the behavioral pattern information contained in the user's past trajectory is not effectively separated, which is unfavorable for learning user's personalized preferences.

[0004] Therefore, how to build a proven model to establish multimodal information relationships and fully integrate them is currently a technical problem that needs to be urgently solved. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the present invention provides a POI recommendation method based on time-series knowledge map, which can effectively solve the above problems.

[0006] The technical solutions specifically adopted by the present invention are as follows: In a first aspect, the present invention provides a POI recommendation method based on a time-series knowledge map, which includes the following steps: S1: Construct a dynamic time series knowledge map and a static group knowledge map based on the complete historical behavior trajectories of all users. The dynamic time series knowledge map is a map set consisting of dynamic relationship knowledge maps of different historical time slices, each of which records the dynamic relationships between all users and points of interest within a historical time slice. The dynamic relationships include access relationships for recording users' access behavior to points of interest and follow relationships for recording users' access behavior to neighboring points of interest. The static group knowledge map records the static relationships between all users and points of interest across all historical time slices. The static relationships include social relationships for recording friendship relationships between users, location relationships for recording spatial regions in which points of interest are located, adjacency relationships for recording whether different points of interest belong to neighboring points, category relationships for recording interest point categories to which points of interest belong, and group relationships for recording users grouped according to the points of interest they visit and the spatial regions they visit. S2: Obtain the historical behavior trajectory substring of the target user before the predicted waiting time, and sequentially extract the user comment text of each interest point accessed by the user from it. Use the aspect-based sentiment analysis module constructed based on the pre-trained model to perform word embedding on the user comment text, and concatenate the sentiment embeddings of all user comment texts to obtain a user comment sentiment embedding sequence. S3: The historical behavior trajectory substring, dynamic time series knowledge map, static group knowledge map and user comment sentiment embedding sequence are input into an interest point recommendation model, and first, an embedding module performs word embedding operation on the input data, and then a multimodal knowledge fusion module performs fusion operation on the dynamic time series knowledge map and the static group knowledge map based on a heterogeneous mutual attention mechanism, and fuses interest points, users and other multimodal information to obtain interest point fusion feature representation and user fusion feature representation, and finally, a decoding module combines the interest point fusion feature representation and the user fusion feature representation, and then inputs them into a cascaded recurrent neural network and a multi-layer sensor to predict the interest points that the target user can access at the next time.

[0007] As a preferred embodiment of the first aspect, in the dynamic time-series knowledge map, the access relationship is recorded by a four-element tuple consisting of a user, an access relationship identifier, an access interest point, and an access time, and the following relationship is recorded by a four-element tuple consisting of a preceding access location, a following relationship identifier, a subsequent access location, and an access time.

[0008] In the static group knowledge map, social relationships are recorded by a three-element tuple consisting of a user, a social relationship identifier, and a user; location relationships are recorded by a three-element tuple consisting of an interest point, a location relationship identifier, and the located Geohash-5 spatial region; adjacent relationships are recorded by a three-element tuple consisting of an interest point, an adjacent relationship identifier, and an interest point; category relationships are recorded by a three-element tuple consisting of an interest point, a category relationship identifier, and the interest point category to which it belongs; and group relationships are recorded by a three-element tuple consisting of a user, a group relationship identifier, and the user group to which it belongs. Here, user groups are divided into two categories: one is an interest point level group obtained by clustering based on the interest points accessed by users, and the other is an area level group obtained by clustering based on the Geohash-5 spatial region to which users belong.

[0009] In the first embodiment, the aspect-based sentiment analysis module is preferably obtained by cascading pre-trained DistilBERT models into a single multi-class classifier and then fine-tuning the overall configuration. The aspect-based sentiment analysis module first generates an embedding for the user comment text using the DistilBERT model, then inputs the embedding into a multi-class classifier to obtain comment dimensions corresponding to the user comment text and positive / negative scores for each comment dimension, and combines the positive / negative scores for all comment dimensions to output a sentiment embedding corresponding to the user comment text.

[0010] Furthermore, the comment dimension includes three dimensions: product, price, and service.

[0011] As a preferred embodiment of the first aspect, the processing flow in the multimodal knowledge fusion module is as follows. S31, the dynamic time series knowledge map and the static group knowledge map are input into the heterogeneous diagram attention network respectively to perform information fusion, and the fused dynamic time series knowledge map and static group knowledge map are obtained. S32, arrange all interest points in the historical behavior trajectory substring in the order of user access, sequentially extract hidden layer vectors corresponding to each interest point from the fused dynamic time series knowledge map, construct a user behavior trajectory embedding with global time slice information, sequentially extract hidden layer vectors corresponding to each interest point from the fused static group knowledge map, construct a user behavior trajectory embedding with global static information, and extract interest point level group features and area level group features from the fused static group knowledge map. S33: Using the user comment emotion embedding sequence as a query, fuse the user behavior trajectory embedding with global time slice information and the original user behavior trajectory embedding through an attention mechanism to obtain a fused user behavior trajectory embedding. The fused user behavior trajectory embedding is used as a value, the user behavior trajectory embedding with global static information is used as a query, and the interest point level group feature is used as a key to input into the Encoder module of the Transformer model for fusion encoding, thereby obtaining an interest point fusion feature representation. S34: The interest point level group features and the area level group features are connected and fused, and the obtained fused group features are used as keys, the user embedded feature representations of all users in the fused static group knowledge map are used as queries, and the user embedded feature representations of all users in the original static group knowledge map are used as values, which are input into the Encoder module of the Transformer model for fusion encoding to obtain the user fused feature representation.

[0012] Preferably, in the first aspect, the interest point recommendation model needs to be optimized in advance by a total loss function obtained by weighting an interest point prediction loss and a static map loss.

[0013] JPEG0007804300000001.jpg62153

[0014] In a second aspect, the present invention provides a computer program product, which includes computer programs / instructions, which, when executed by a processor, can realize the method for recommending points of interest based on a time-series knowledge map according to any one of the embodiments of the first aspect.

[0015] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored therein, which, when executed by a processor, can realize the method for recommending points of interest based on a time-series knowledge graph according to any one of the embodiments of the first aspect.

[0016] In a fourth aspect, the present invention provides a computer electronic device, the electronic device including a memory and a processor.

[0017] The memory is used to store computer programs.

[0018] When the processor executes the computer program, it can realize the method for recommending points of interest based on a time-series knowledge graph according to any one of the embodiments of the first aspect.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a method for recommending next points of interest based on a time-series knowledge map for multimodal scenes. The method constructs a dynamic time-series knowledge map and a static group knowledge map based on the user's historical action trajectory, respectively, and learns the user's dynamic action preferences and static features. The present invention utilizes a heterogeneous mutual attention mechanism to aggregate information from the dynamic time-series knowledge map and the static group knowledge map, learning the heterogeneous information in the knowledge maps. Furthermore, by introducing a multimodal knowledge fusion module, cross-learning is performed between user features and interest point features with the assistance of high-quality semantic information reflecting user emotional trends extracted from user comment text by an aspect-based sentiment analysis module, effectively addressing the information fusion problem in real scenes and providing guidance for next point of interest recommendation. The present invention has the characteristics of high accuracy and strong scalability, and can quickly grasp the direction of user actions, providing technical support for personalized user action trajectory prediction. [Brief explanation of the drawings]

[0020] [Figure 1] 1 is a schematic diagram illustrating steps of a method for recommending points of interest based on a time-series knowledge map in an embodiment of the present invention; [Figure 2] FIG. 1 is a schematic diagram of a network architecture of an interest point recommendation model in an embodiment of the present invention; [Figure 3] FIG. 2 is a schematic diagram of a dynamic time-series knowledge map according to an embodiment of the present invention. [Figure 4] FIG. 1 is a schematic diagram of a dynamic time-series knowledge map of user B as an example and a user behavior trajectory embedding extracted therefrom in an embodiment of the present invention. [Figure 5] 1 is a schematic diagram of a configuration of a computer electronics device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0021] In order to more clearly understand the objectives, technical solutions and advantages of the present invention, the present invention will be described in more detail below with reference to the accompanying drawings. The present invention is defined by the claims, and covers any alternatives, modifications, equivalent methods and means that fall within the spirit and scope of the present invention. Furthermore, in order to provide a better understanding of the present invention to the public, certain details are set forth in the following detailed description of the present invention. However, those skilled in the art will be able to fully understand the present invention without these details. In describing the present invention, the terms "first" and "second" are used merely for descriptive purposes and should not be construed as implying the relative importance or quantity of the indicated technical features. Thus, a feature that includes "first" and "second" may explicitly or implicitly include at least one such feature.

[0022] In a preferred embodiment of the present invention, as shown in FIG. 1, the method for recommending points of interest based on the time-series knowledge map is specifically implemented by the following steps S1 to S3. The method for recommending points of interest of the present invention uses a deep learning method to build a network framework based on the user's past check-in data and multimodal information, extracts the user's preference features as shown in FIG. 2, and accurately predicts the places the user is currently most likely to visit. The present invention uses a dynamic time-series knowledge map and a static group knowledge map to learn and organically integrate trajectory sequences and multimodal relationships. At the same time, the dynamic time-series knowledge map effectively separates the behavior mode information contained in the user's historical trajectory, and objectively learns the inherent modes and evolutionary relationships of each time slice to realize next point of interest prediction that matches the user's personalized preferences.

[0023] A specific method for implementing each of steps S1 to S3 will be described in detail below. S1: Construct a dynamic time series knowledge map and a static group knowledge map based on the complete historical behavior trajectories of all users. The dynamic time series knowledge map is a map set consisting of dynamic relationship knowledge maps of different historical time slices, each of which records the dynamic relationships between all users and points of interest within a historical time slice. The dynamic relationships include access relationships for recording users' access behavior to points of interest and follow relationships for recording users' neighboring access behavior to different points of interest. The static group knowledge map records the static relationships between all users and points of interest across all historical time slices. The static relationships include social relationships for recording friendship relationships between users, location relationships for recording spatial regions in which points of interest are located, adjacency relationships for recording whether different points of interest belong to neighboring points, category relationships for recording interest point categories to which points of interest belong, and group relationships for recording users' grouping according to the points of interest they visit and the spatial regions they visit.

[0024] The construction of the two types of maps requires the complete historical behavioral trajectories of all users, which can be obtained from social platforms that record users' access behavior to POIs, such as Foursquare, Gowalla, and Yelp. The construction of the two types of maps requires the trajectories of all users on the social platforms, and the time span of one complete historical behavioral trajectory is a specified historical period. The specific length of the historical period can be reasonably selected based on the actual data situation, for example, the most recent six months or the most recent year.

[0025] The specific method for constructing a knowledge map belongs to the prior art, and the relationships between entities can be recorded by rationally designed tuples.

[0026] In an embodiment of the present invention, for two types of dynamic relationships existing in the dynamic time-series knowledge map, the access relationship can be recorded by a four-element tuple consisting of a user, an access relationship identifier, an access interest point, and an access time, and the following relationship can be recorded by a four-element tuple consisting of a preceding access location, a following relationship identifier, a subsequent access location, and an access time.

[0027] JPEG0007804300000002.jpg127155

[0028] In an embodiment of the present invention, constructing a static group knowledge map simply requires constructing all data over a complete time span as a single map, without the need to consider different historical time slices as in constructing a dynamic time-series knowledge map. The five types of static relationships present in the static group knowledge map include: social relationships recorded by a three-element tuple consisting of a user, a social relationship identifier, and a user; location relationships recorded by a three-element tuple consisting of a point of interest, a location relationship identifier, and the geohash-5 spatial domain where the user is located; adjacent relationships recorded by a three-element tuple consisting of a point of interest, an adjacent relationship identifier, and a point of interest; category relationships recorded by a three-element tuple consisting of a point of interest, a category relationship identifier, and the associated point of interest category; and group relationships recorded by a three-element tuple consisting of a user, a group relationship identifier, and the associated user group. User groups are divided into two categories: interest point-level groups obtained by clustering based on the points of interest accessed by users; and area-level groups obtained by clustering based on the geohash-5 spatial domains accessed by users. The cluster division here essentially divides users into groups based on the interest points and Geohash-5 spatial regions they access, i.e., users who access the same interest point are placed in the interest point level group for that interest point, and users who access the same Geohash-5 spatial region are placed in the region level group for that Geohash-5 spatial region.

[0029] JPEG0007804300000003.jpg112153

[0030] The Geohash-5 spatial region used in the present invention is obtained by encoding the entire geographic space using the Geohash algorithm. This algorithm is an address encoding method that can encode two-dimensional spatial latitude and longitude data into a single string. This algorithm belongs to the prior art and can be calculated directly using conventional correlation functions or programs. The basic steps of the algorithm are as follows: first, convert the longitude and latitude into binary according to different accuracy requirements, then combine the longitude and latitude, with the longitude and latitude occupying even digits and odd digits, respectively, and finally encode the binary string using Base32. The longer the encoding, the smaller the display range and the more accurate the location. The specific value of the accuracy can be optimized according to actual practice. In this embodiment of the present invention, the accuracy is set to 5, that is, the encoding length is set to 5.

[0031] The dynamic time-series knowledge map is constructed based on user entities, interest point entities, and the dynamic relationships between them, and can be used to learn user behavior patterns and time-varying behavioral preferences in each time slice. The static group knowledge map is constructed based on static relationships between entities, and can be used to learn multi-dimensional heterogeneous information and stable feature dependencies that do not change over time. Both can provide user preference information for future interest point selection from different dimensions.

[0032] S2: Obtain the historical behavior trajectory substring of the target user before the predicted waiting time, and sequentially extract the user comment text of each interest point accessed by the user from it. Use the aspect-based sentiment analysis module constructed based on the pre-trained model to perform word embedding on the user comment text, and concatenate the sentiment embeddings of all user comment texts to obtain a user comment sentiment embedding sequence.

[0033] It should be noted that the historical behavior trajectory substring before the predicted waiting time of the target user refers to a behavior trajectory consisting of a series of interest points recently accessed by the target user who needs to make interest point recommendations before the predicted waiting time. The number of interest points constituting the historical behavior trajectory substring can be adjusted according to actual needs. In the embodiment, the most recent 20 interest points can be adopted, that is, the length of the historical behavior trajectory substring is 20.

[0034] Theoretically, the above-mentioned aspect-based sentiment analysis module can be trained and fine-tuned based on any pre-trained language model. In an embodiment of the present invention, considering the requirements for model size and execution speed in actual implementation scenarios, the aspect-based sentiment analysis module is preferably constructed using a DistilBERT model pre-trained on a large corpus. After the pre-trained DistilBERT model, a multi-classifier needs to be cascaded. The cascaded models are then fine-tuned together on a sentiment analysis dataset to obtain the aspect-based sentiment analysis module. The processing flow of the aspect-based sentiment analysis module is as follows: First, an embedding representation is generated for the input user comment text using the DistilBERT model. Then, the embedding representation is input to the multi-classifier to obtain the comment dimension corresponding to the user comment text and the positive / negative score (a two-dimensional vector recording positive and negative scores) for each comment dimension. The positive / negative scores for all comment dimensions are concatenated and output as a sentiment embedding corresponding to the user comment text.

[0035] The sentiment analysis dataset used in the fine-tuning process is obtained by manually labeling for the task of the present invention. The training samples in the sentiment analysis dataset include user comment texts for points of interest, as well as truth-value labels for the comment dimensions and positive / negative comment scores of the user comment texts. After cascading the DistilBERT model and a multi-classifier, supervised learning is performed on the sentiment analysis dataset, and fine-tuning is completed after convergence. The specific comment dimensions can be designed according to the actual situation of the collected user comment text data. For example, on a typical review website, the dimensions related to user comment texts for points of interest cover three dimensions: product, price, and service. These three dimensions can be considered as the comment dimensions output by the multi-classifier.

[0036] The product in the comment dimension mentioned above refers to a service product provided by a merchant corresponding to a point of interest; for example, for a restaurant, the product is food, and for an amusement park, the product is an attraction.

[0037] In an embodiment of the present invention, the entire process of pre-training and fine-tuning may be specifically implemented by the following steps. S21,We pretrain the DistilBERT pretrained model on large-scale corpora, namely, the BookCorpus dataset and the English Wikipedia data.

[0038] JPEG0007804300000004.jpg82153

[0039] JPEG0007804300000005.jpg26153

[0040] JPEG0007804300000006.jpg54153

[0041] S3: The target user's historical behavior trajectory substring before the predicted waiting time, the dynamic time series knowledge map, the static group knowledge map and the user comment emotion embedding sequence are input into the interest point recommendation model. First, the embedding module performs a word embedding operation on the input data. Then, the multimodal knowledge fusion module performs a fusion operation on the dynamic time series knowledge map and the static group knowledge map based on the heterogeneous mutual attention mechanism, and fuses the interest points, user and other multimodal information to obtain an interest point fusion feature representation and a user fusion feature representation. Finally, the decoding module combines the interest point fusion feature representation and the user fusion feature representation, and then inputs them into a cascaded recurrent neural network and a multi-layer sensor to predict the interest points that the target user can access at the next time.

[0042] JPEG0007804300000007.jpg40154

[0043] In an embodiment of the present invention, the multimodal knowledge fusion module performs fusion operations on the dynamic time-series knowledge map and the static group knowledge map respectively, and can fuse POIs, users and other multimodal information to obtain richer context information. The processing flow in the multimodal knowledge fusion module is as follows:

[0044] JPEG0007804300000008.jpg63154

[0045] JPEG0007804300000009.jpg156154

[0046] S33: Using the user comment emotion embedding sequence as a query, fuse the user behavior trajectory embedding with global time slice information and the original user behavior trajectory embedding through an attention mechanism to obtain a fused user behavior trajectory embedding. The fused user behavior trajectory embedding is used as a value, the user behavior trajectory embedding with global static information is used as a query, and the interest point level group feature is used as a key to input into the Encoder module of the Transformer model for fusion encoding, thereby obtaining an interest point fusion feature representation.

[0047] The query, key, and value in the present invention are Q (Query), K (Key), and V (Value) in the attention mechanism, respectively.

[0048] JPEG0007804300000010.jpg96155

[0049] JPEG0007804300000011.jpg55155

[0050] S34: The interest point level group features and the area level group features are connected and fused, and the obtained fused group features are used as keys, the user embedded feature representations of all users in the fused static group knowledge map are used as queries, and the user embedded feature representations of all users in the original static group knowledge map are used as values, which are input into the Encoder module of the Transformer model for fusion encoding to obtain the user fused feature representation.

[0051] JPEG0007804300000012.jpg64154

[0052] JPEG0007804300000013.jpg80154

[0053] JPEG0007804300000014.jpg62154

[0054] JPEG0007804300000015.jpg78153

[0055] JPEG0007804300000016.jpg24153

[0056] The specific model training process belongs to the prior art. By combining the total loss function and the optimizer, the learnable parameters can be continuously optimized, and the interest point recommendation model after optimization can be used for actual inference.

[0057] It should be noted that the method steps shown in S1 to S3 above can be essentially realized in the form of a computer program.

[0058] Based on the same inventive concept as above, as shown in FIG. 5, the present invention further provides a computer electronic device corresponding to the interest point recommendation method based on time-series knowledge graph provided in the above embodiment, where the electronic device includes a memory and a processor.

[0059] The memory is used to store computer programs.

[0060] When the processor executes the computer program, it is used to implement the above-mentioned method for recommending points of interest based on a time-series knowledge graph.

[0061] Furthermore, the logic instructions in the memory can be realized in the form of a software functional unit and stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the essence of the technical solution of the present invention or a contribution to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (such as a personal computer, a server, or a network device) to perform all or some of the steps of the method described in each embodiment of the present invention.

[0062] Based on the same inventive concept as above, the present invention provides a computer-readable storage medium corresponding to a POI recommendation method based on a time-series knowledge graph, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned point of interest recommendation method based on a time-series knowledge graph can be realized.

[0063] Based on the same inventive concept as above, the present invention provides a computer program product, which includes computer programs / instructions, and when the computer programs / instructions are executed by a processor, can realize the above-mentioned method for recommending points of interest based on a time-series knowledge graph.

[0064] Specifically, in the computer-readable storage medium of the three embodiments, when the stored computer program is executed by a processor, the above steps S1 to S3 can be executed.

[0065] As can be understood, the storage medium can include a random access memory (RAM) and a non-volatile memory (NVM), such as at least one disk memory, and can also be various media capable of storing program code, such as a USB memory, a portable hard disk, a magnetic disk, or an optical disk.

[0066] As can be appreciated, the processor may include a general-purpose processor, such as a Central Processing Unit (CPU), a Network Processor (NP), etc. It may also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component.

[0067] For the sake of convenience and brevity, those skilled in the art may refer to the corresponding processes in the aforementioned method embodiments for the specific operating processes of the system described above, and the description will not be repeated here. In each embodiment provided in this application, the division of steps or modules in the system and method is merely a logical functional division, and other division methods may be used in actual implementation. For example, multiple modules or steps may be combined or integrated, and a single module or step may be further divided.

[0068] In order to facilitate understanding of the essence of the present invention, the detailed implementation process and technical effects of the interest point recommendation method based on the time-series knowledge map shown in S1 to S3 above on a specific dataset will be shown below through more specific examples.

[0069] [Example] The steps of this embodiment are the same as the method for recommending points of interest based on the time-series knowledge map shown in the above-mentioned steps S1 to S3, and will not be repeated here, but will mainly show the specific data set, some specific parameter settings, and implementation results of this embodiment.For the sake of convenience, the method abbreviated as steps S1 to S3 below will be referred to as the method of the present invention, and the used model for recommending points of interest will be referred to as MINet.

[0070] The original data used in this study are three widely used real-world scene datasets: Foursquare, Gowalla, and Yelp. The Gowalla dataset contains 3,300,986 check-in events from 52,979 users at a total of 121,851 locations. The Foursquare dataset contains 9,447,873 check-in events from 46,065 users at a total of 69,005 locations. The Yelp dataset contains 632,476 check-in events from 9,627 users at a total of 8,696 locations. Because the Foursquare and Gowalla datasets do not contain user comment information, we also constructed a default version of the MINet model without inputting user comment sentiment embedding sequences. This model, named MINet-Revised, was used as a control method in subsequent experiments.

[0071] JPEG0007804300000017.jpg31152

[0072] Furthermore, the experiment in this example compares the method of the present invention with several conventional prediction methods. The conventional prediction methods used for comparison are as follows: (1) FPMC: a conventional Markov chain model based on user personalized behavior; (2) RNN: a recurrent neural network for prediction based on historical trajectory sequences; (3) DeepMove: a recurrent neural network based on the fusion of attention information and trajectory sequences; (4) STAN: a two-layer attention network based on spatio-temporal attention; (5) TiSASRec: an attention network based on sequence position and time interval; (6) Flashback: a recurrent neural network based on context information in past and current hidden layers; (7) GETNext: a Transformer model that predicts user behavior based on global transition probability; (8) a model that learns POI transition relationships based on spatio-temporal knowledge maps; and (9) MARAN: a model based on aggregation of local centroid trajectories and user behavior patterns. In this example, mean average precision (Acc@K) and mean reciprocal rank (MRR) are used as evaluation metrics for the prediction models. Acc@K calculates the ratio of true positive samples among the top K predicted samples. In our experiments, K = {5, 10}. MRR can reflect the overall performance of the recommendation and places more importance on the prediction ranking.

[0073] The final experimental results, as shown in Table 1, demonstrate that MINet, the method of the present invention, achieved better results than the control model on the Yelp dataset. Specifically, on the Yelp dataset, MINet achieved improvements of 9.86%, 6.48%, and 7.21% compared to the best-performing control model, MARAN, in the Acc@5, Acc@10, and MRR metrics, respectively. The default version of MINet-Revised also achieved better results than the control model on Gowalla, with improvements of 2.56%, 0.90%, and 3.45%, respectively. On the Foursquare dataset, MINet performed slightly worse than the best-performing control model, MARAN, with an average difference of 1.40%. In particular, on the Yelp dataset, MINet achieved improvements of 47.17%, 45.24%, and 33.53% compared to the default version in each metric, demonstrating the effectiveness of the method of the present invention. [Table 1] Comparison of experimental results between the method of the present invention and the control method JPEG0007804300000018.jpg89165

[0074] The above embodiment is a preferred solution of the present invention and does not limit the present invention. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by adopting the method of equivalent substitution or equivalent transformation is included in the protection scope of the present invention.

Claims

1. A computer program implemented in the form of a computer program, the computer program being executed by a processor as instructions for executing a method for recommending points of interest (POI) based on a time-series knowledge map executed by a computer electronics device, comprising: S1: constructing a dynamic time series knowledge map and a static group knowledge map based on the complete historical behavior trajectories of all users, the dynamic time series knowledge map being a map set consisting of dynamic relationship knowledge maps of different historical time slices, each dynamic relationship knowledge map recording the dynamic relationships between all users and points of interest in the historical time slice, the dynamic relationships including access relationships for recording the users' access behavior to points of interest and follow relationships for recording the users' neighboring access behavior to different points of interest; the static group knowledge map recording the static relationships between all users and points of interest in all historical time slices, the static relationships including social relationships for recording the friendship relationships between users, location relationships for recording the spatial regions in which points of interest are located, adjacent relationships for recording whether different points of interest belong to neighboring points, category relationships for recording the interest point categories to which points of interest belong, and group relationships for recording the users' grouping according to the points of interest accessed and the spatial regions accessed; S2. Obtain the historical behavior trajectory substring of the target user before the predicted waiting time, and sequentially extract the user comment text of each interest point accessed by the user from it. Use the aspect-based sentiment analysis module constructed based on the pre-trained model to perform word embedding on the user comment text, and concatenate the sentiment embeddings of all the user comment texts to obtain a user comment sentiment embedding sequence; S3, the historical behavior trajectory substring, the dynamic time series knowledge map, the static group knowledge map and the user comment sentiment embedding sequence are input into an interest point recommendation model, and the embedding module first performs a word embedding operation on the input data, and then the multimodal knowledge fusion module performs a fusion operation on the dynamic time series knowledge map and the static group knowledge map based on a heterogeneous mutual attention mechanism, and fuses the interest point, user and multimodal information to obtain an interest point fusion feature representation and a user fusion feature representation, and finally the decoding module combines the interest point fusion feature representation and the user fusion feature representation, and then inputs them into a cascaded recurrent neural network and a multi-layer sensor to predict the interest points that the target user can access at the next time; the multimodal information includes dynamic relations including the access relations and the following relations, and static relations including the social relations, the location relations, the adjacent relations, the category relations, and the group relations; The process of S2 is specifically carried out by the following steps: S21. Pre-train the DistillBERT pre-trained model on large-scale corpora, namely the BookCorpus dataset and English Wikipedia™ data; S22. Classify the comment dimensions of the user comment text z into three indicators: product, price, and service; use the DistillBERT pre-trained model to obtain the embedding representation of the user comment text z; calculate the classification score for each comment dimension and the positive / negative score of the user comment text for each comment dimension; S23, connect the positive / negative scores of each comment dimension to obtain the sentiment embedding corresponding to the user comment text; S24. Combine the classification scores and the truth labels of the positive / negative scores, and use the cross-entropy loss function to obtain a final total loss function, thereby optimizing the learnable parameters in DistillBERT and multi-classifiers. A computer program characterized by:

2. A computer-readable storage medium, 2. A storage medium storing the computer program according to claim 1.

3. 1. A computer electronic device comprising: The electronic device includes a memory and a processor; The memory is used to store the computer program of claim 1, The processor is a computer electronic device that can read and execute the computer program from the memory.

Images