Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/hindawi-publishing-corporation/personalized-travel-route-recommendation-model-of-intelligent-service-56T3SiViDz
References
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
- Publisher
- Hindawi Publishing Corporation
- Copyright
- Copyright © 2022 Xiang Huang. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
- ISSN
- 1687-9600
- eISSN
- 1687-9619
- DOI
- 10.1155/2022/7778592
- Publisher site
- See Article on Publisher Site
Abstract
Hindawi Journal of Robotics Volume 2022, Article ID 7778592, 8 pages https://doi.org/10.1155/2022/7778592 Research Article Personalized Travel Route Recommendation Model of Intelligent Service Robot Using Deep Learning in Big Data Environment Xiang Huang Hunan Mass Media Vocational and Technical College, Changsha, Hunan 410100, China Correspondence should be addressed to Xiang Huang; 332374935@qq.com Received 8 December 2021; Accepted 15 January 2022; Published 29 January 2022 Academic Editor: Shan Zhong Copyright © 2022 Xiang Huang. -is is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Aiming at the problems that the traditional model is difficult to extract information features, difficult to learn deep knowledge, and cannot automatically and effectively obtain features, which leads to the problem of low recommendation accuracy, this paper proposes a personalized tourism route recommendation model of intelligent service robot using deep learning in a big data environment. Firstly, by crawling the relevant website data, obtain the basic information data and comment the text data of tourism service items, as well as the basic information data, and comment the text data of users and preprocess them, such as data cleaning. -en, a neural network model based on the self-attention mechanism is proposed, in which the data features are obtained by the Gaussian kernel function and node2vec model, and the self-attention mechanism is used to capture the long-term and short-term preferences of users. Finally, the processed data is input into the trained recommendation model to generate a personalized tourism route recommendation scheme. -e experimental analysis of the proposed model based on Pytorch deep learning framework shows that its Pre@10, Rec@10 values are 88% and 83%, respectively, and the mean square error is 1.537, which are better than other comparison models and closer to the real tourist route of the tourists. recommendations that meet their specific needs to help users 1. Introduction quickly filter useless information in a large amount of travel With the continuous improvement of social living standards, information and improve the efficiency and comfort of users in integrating information [5]. -e personalized route people’s demand for tourism and leisure is also increasing year by year. -e development and prosperity of the tourism recommendation platform is diversified. Among them, in- industry have made going out to travel increasingly popular. telligent service robots in scenic spots occupy an important In the preparation of outbound travel, the route formulation position among many platforms because of their high ef- in the travel strategy is an extremely important and critical ficiency and convenience. -erefore, its recommended al- step [1]. People can check relevant travel guides or collect gorithm model is also very important. relevant information on the internet, however, it will waste -e purpose of personalized travel route recommen- time and be inefficient. Moreover, the information they find dation is to recommend a travel route composed of multiple will not match their current needs [2]. -erefore, it is dif- points of interest (POI) based on the user’s personalized ficult to find information that fits the purpose, and sys- interests and the user’s own travel restrictions. In the per- tematically, there will be big differences between the sonalized recommendation of POI, the key is to compre- hensively model the user interests and similarities between information provided by different people. In addition, the rapid increase in the number of users of social networks has the users. A comprehensive analysis of the user’s person- caused a rapid increase in network information. When users alized preference for different POIs and the degree of face massive network data, they cannot realize a quick se- similarity correlation between each POI is used to determine lection of information [3, 4]. -erefore, the author hopes to the user’s personalized interest [6]. In addition, similarity be able to automatically obtain personalized travel matching is used to find the degree of association between 2 Journal of Robotics However, the accuracy rate of the optimal travel route needs the users. By the analysis of similar users and similar mobile patterns, personalized POI recommendations are made to to be improved. Reference [15] proposed a route recom- mendation method based on interest topic and distance users. At the same time, personalized travel route recom- mendation also needs to take into account the user’s per- matching, which obtains the best travel path by analyzing the sonalized factors and generate a travel route that meets their user’s real historical travel footprint and scenic spot resi- own travel restrictions for each user [7]. dence time and combined with the given travel time limit. -ere have been many pieces of research on personalized However, this method has poor timeliness and adaptability travel route recommendation at home and abroad. Refer- and cannot be applied to the independent intelligent robot ence [8] introduced the evolution process of the travel platform. Based on the above analysis, the traditional tourism recommendation system in detail and conducted research on its characteristics and current limitations. At the same route recommendation model is difficult to pay attention to the long-term preferences of users and the poor recom- time, the key algorithms used in the classification and recommendation process and the indicators that can be used mendation effect caused by sparse data. -is paper proposes a personalized travel route recommendation method for to evaluate the performance of the algorithms are also discussed. In terms of recommendation methods, the most intelligent service robots using deep learning in a big data common method is the recommendation technology based environment. It can be applied to intelligent service robots on collaborative filtering. Realize the personalized recom- placed in the halls of scenic spots to realize a personalized mendation by mining the similarity between the users. travel route recommendation. To fuse multisource hetero- Reference [9] uses the sequential pattern mining algorithm geneous data, the proposed model uses the Gaussian kernel to generate various fine-grained candidate POI routes from function, node2vec model, and other technologies to con- struct the embedded representations of users, time, space, POI access sequences to realize the recommendation of the best tourism route. However, the overall recommendation POI score, access frequency, and social relationships. Send it to the deep learning network for analysis. It solves the efficiency is low, and it is slightly insufficient for route planning with complex and multiple points of interest. problem of low recommendation accuracy caused by sparse data. -e experimental results based on the pytoch deep Reference [10] proposed a personalized and content-adap- tive cultural heritage route recommendation to achieve the learning framework show that the proposed model inte- best cultural heritage experience of context-aware routes. grates user preference characteristics, geographical factor Use the first-order Markov model to convert motion as the characteristics, and theme factor characteristics and can time of the problem to realize route recommendation. -e better complete tourism route recommendation, with a overall recommendation effect is good, however, it takes a Pre@5 of 95%. long time and is not suitable for immediate recommenda- tion. Reference [11] proposed a new travel route mining 2. POI Recommended Problem Description method on the basis of considering the theme level and characteristics of scenic spots, in which the scenic spots are -e user “sign-in” record data in social networks contains a subject layered according to the location information of the large amount of high-value information data about POI and popular scenic spots. -e travel path data set is constructed, user preferences, which provides an opportunity for in- and the travel routes are mined in combination with the depth research on personalized POI recommendations. subject level, however, there are still deficiencies in the However, in practical applications, there are some person- consideration of user interest point matching. alized differences in users’ preferences for POI categories With the rapid increase in the demand for tourist route [16]. -e existing POI recommendation methods are mostly recommendation in recent years, thanks to the rapid ad- implemented by content-based or model-based collabora- vancement of computer technology and communication tive filtering technology. -e subject of POI and the rela- technology, machine learning technology has been widely tionship between the subjects are not fully considered. used in the field of automatic recommendation of demand -erefore, in the user’s personalized recommendation, [2]. Among them, deep learning algorithms have achieved combined with the theme factors of POI, more effective excellent results in many fields. It can effectively process features are obtained from the limited user access infor- unstructured multimedia data. Some scholars have begun to mation, and appropriate models are selected to achieve try to use convolutional neural networks to solve the feature distinguishable user preference modeling. -ese are the keys engineering problems faced by the recommender systems to improve the effectiveness of personalized POI recom- [12]. Reference [13] proposed a matrix factorization algo- mendations [17]. rithm based on two-stage clustering. Using the social net- Finding effective features from the check-in data is the work subgraph integrated with preference similarity scores, key to improve the quality of POI recommendations. Tra- combined with geographic spatial influence, the cluster ditional methods only learn the linear or low-order inter- refinement of preference embeddings is extended to the action between the features, and they cannot effectively cluster refinement of geographic preference embeddings. In integrate the features in a location-based social network this way, the best route recommendation under complex (LBSN) [18]. In recent years, with the rapid development of conditions is realized. Reference [14] proposed a person- deep learning, it can intelligently learn high-order charac- alized travel recommendation scheme based on a weighted teristics and interact from the input of specific tasks. multi-information constraint matrix factorization scheme. -erefore, a deep neural network recommendation Journal of Robotics 3 framework that combines the DNN network with the LDA Preprocess data Data collection Data cleaning topic model and matrix factorization algorithm is proposed, named DLM. -e user preference feature, geographic factor feature, and probability topic feature in LBSN are integrated into the POI recommendation task using word-embedding Construction of depth technology. High-level interactions between the features are Network construction prediction model learned through neural networks, and personalized rec- ommendations are made to users. Train the constructed 3. Proposed Model Network training network 3.1. Overall Framework. -e proposed implementation framework of the recommended model is mainly divided into four modules. -ey are data preprocessing, deep pre- Generate Recommended results diction model construction, network training, and final recommendation list recommendation list generation. -e main implementation framework is shown in Figure 1. -e model preprocesses the Figure 1: Main implementation framework based on depth pre- acquired data and uses the Gaussian kernel function and diction model recommendation. node2vec model to model it to obtain the corresponding POI location and social relationship embedded representation. historical comment items. -e comment data mainly in- Both of them are input into the self-attention module to cludes the user’s comment text information on past travel capture user preferences to obtain an ideal personalized service items and corresponding scores. -is part of the tourism recommendation scheme. content is mainly used to extract user behavior character- -e process of data acquisition and preprocessing is very istics, analyze user preferences, and build user characteristic complex. Mainly by crawling the relevant website data, the models. basic information data, the comment text data of travel -e second part is the basic information of tourism service items, and the basic information data and comment service items and the comment data of tourism service items. text data of users are obtained. -en, these data are pre- -e basic information of the travel service item includes the processed. Perform data cleaning on the crawled data to name, location, and label of the travel service item. -e filter out the incomplete data and junk data. comment data mainly refers to the comment text infor- mation and the corresponding score obtained by the tourism service item. -is part of the data is used to extract the 3.2. Construction of Deep Prediction Models. -e construc- attribute characteristics of the tourism service items and tion of in-depth prediction models. Neural network tech- construct the characteristic model of the tourism service nology is mainly used to construct a network model and items. process the preprocessed data. Use the feature extraction To obtain these data, web crawlers are used to crawl the ability of deep learning to obtain the corresponding features, related travel websites. -e crawler uses the Scrapy web and use the model to predict the user's rating of the tourism crawler framework to crawl the website. Scrapy is a dis- service item [19]. tributed crawler framework based on Python. Scrapy is Train the network. For the constructed model network, highly flexible and controllable and can easily implement use the training sample data for supervised network training. distributed crawlers. At the same time, Scrapy encapsulates Mine the potential factors between the users and tourism the implementation details of a lot of crawlers, which can service items and learn the expression of the interaction focus more on data extraction. relationship between the users and tourism service items to train the model. Generate a personalized recommendation list. Test the 3.3.2. Data Cleaning. To ensure that the recommended experimental data and input the experimental data into the results are valid, the data should be complete and reliable. trained model. -e model predicts the user’s rating of travel -erefore, the crawled data must first be cleaned up to filter service items and sorts them according to the size of the out the incomplete data and junk data. During data cleaning, rating. Generate a personalized recommendation list for the steps followed will be as follows: each user to complete the user’s recommendation. (1) Firstly, filter out the users with incomplete basic information. Incomplete basic information means 3.3. Data Acquisition and Data Preprocessing that the basic information characteristics cannot be found. For tourism service items and users, it is 3.3.1. Data Collection. -e collected data mainly includes impossible to dig out the characteristic influence of two parts. its basic information. -erefore, ensuring the in- -e first part is the user’s basic information and the tegrity of basic information plays an important role comment text data. -e user’s basic information data mainly in model building. includes the user’s gender, age, occupation, city, and RELU Embedded layer 4 Journal of Robotics (2) Spam comments need to be filtered. By observing the Subsidiary information embedding scraped comment data, it can be found that the RELU general comment has only one word, and words that do not indicate good or bad mood can be filtered out as spam comments as these data have no positive RELU effect on the establishment of the model. (3) To filter the content of the comments, filter out the RELU special symbols in the comments. -ese symbols are not helpful in digging out the characteristics of the comment content. After the data cleaning is com- pleted, complete and reliable data is obtained. -ese data will be further processed. User embedding 3.4. Network Building. -e proposed model contains two Input data components: information embedding and information in- teraction. Its structure is shown in Figure 2. For the in- Self-Attention formation embedding module, firstly, the input data is FC multi-hot coded, and the user POI check-in sequence model is constructed to generate the potential representation matrix. -en, in the auxiliary information extraction part, POI geographic location information is extracted by the Output Sigmoid Gaussian kernel function, and the data information is Figure 2: Overall framework of the proposed method. normalized by softmax. In the information interaction module, the deeper interaction of data is obtained based on self-attention to learn the long-term and short-term pref- � � � �2 � � erences of the users, and the information is fused using three � � L − L � � t t i j ⎜ ⎟ ⎛ ⎜ ⎞ ⎟ ⎜ ⎟ ⎝ ⎠ κ L , L � exp − . (2) bottleneck layers to obtain the final prediction. t t i j 2σ 3.4.1. User Representation Embed. -e purpose of a given Among them, L and L are the geographic coordi- t t user’s check-in record is to learn the potential representation i i nates of the two POIs visited by the user. -e value range of the POI sequence to improve the accuracy of POI rec- of the Gaussian kernel is κ(L , L ) ∈ [0, 1]. σ is the t t ommendations. In the proposed model, a transition matrix i j bandwidth, which controls the radial range of action. is designed. In this way, the user’s spatiotemporal intentions Finally, by calculating the paired Gaussian kernel value of and the potential characteristics of POI are mapped in the each POI pair, the Gaussian kernel value vector κ ∈ R can feature space. -e input is a user sign-in vector u u u u be obtained. � p , p , . . . , p characterized by multi-hot. When i t t t 1 2 i−1 p is 1, it means that user u has visited POI at time t . -e p k mapping process is as follows: 3.4.3. POI Score and Access Frequency Embedding. By u u φ � f ω p + b , (1) 1 u u preprocessing the LBSN data set, users’ ratings and access frequency can be obtained. Use normalization to scale its where φ represents the latent representation vector of user value to (0, 1). -e sum of all check-in score probabilities u. ω and b are the weight and bias of user u, respectively. u u and access frequency probabilities of each user is 1, re- spectively. -erefore, it is easier to characterize the user’s preference for the POI they have visited, which is expressed 3.4.2. Spatio-Temporal Information Embedding. In the as follows: user’s check-in record, the user’s behavior is usually limited to several specific areas, which is a well-known geographical exp L cluster phenomenon in the user’s check-in activities [20]. ϕ L |L � , r j i From this phenomenon, it can be inferred that the users expL j�1 prefer to visit the unreachable POI near the POI they have (3) visited before, and users’ liking depends on the attributes exp L ϕ L |L � . and distance of the two POIs [21, 22]. If the attributes are the q j i expL j�1 same, then the closer the distance between the two POIs, the greater the possibility that the user will visit the unvisited By calculating the score probability value ϕ (L |L ) and POI [23]. To combine the geographical distance attribute of r j i the POI, the Gaussian kernel function is used to extract the the access frequency probability value ϕ (L |L ) of each user q j i separately, the probability value vectors φ ∈ R and neighbor perception influence of the sign-in POI, which is expressed as follows: φ ∈ R can be obtained. q Journal of Robotics 5 3.4.4. User Social Relationship Embedding. Select the Similarly, a user will also have multiple comments, and each node2vec model to extract the user relationship features. For comment has a corresponding user and scenic spot identity each user, use the node2vec model to generate m random (ID) and corresponding score. Because of the limited size of walk sequences of length n. -en, train by Skip-Gram with the table, there is no detailed display here (the comments in hierarchical Softmax. Finally, find users related to the the table only show the content of the comment text). current user, so that a latent representation matrix s ∈ R of -e proposed model is run through the pytorch 1.2.0 the user’s social relationship can be obtained. framework, where the dropout rate is set to 0.5. 4.1. Evaluating Indicator. Two broad indicators are used to 3.4.5. Attached Information Display. To handle the complex evaluate the performance of different recommendation interaction between the user and POI and further POI models, namely accuracy and recall (these two indicators are recommendations, the various auxiliary information is in- represented by Pre@N and Rec@N, respectively), and the tegrated. It is expressed as follows: calculation is as follows: φ � f ReLU κ⊙ φ ⊙ ReLUφ ⊙ s, (4) 2 r q 1 Top − N∩ K Pre@N � , ϑ N where φ is the latent representation vector of ancillary in- u∈U (6) formation. ⊙ is the element dot product. 1 Top − N∩ K Rec@N � , ϑ K u∈U 3.5. Network Training. Use pairwise optimized Bayesian Personalized Ranking (BPR) loss function to learn model where ϑ represents the number of users. N represents the parameters. -e objective function is as follows: number of recommended POIs. Top-N represents the list of the first N points of interest recommended by the recom- Loss � −ln δy − y + λ‖Φ‖ , uL uL 2 i j (5) mendation model to the target user. K represents the real u,L ,L ∈O i j check-in list in the user test set, i.e., the POI set that the user has actually accessed in the actual historical access record. + − where O � (u, L , L )|(u, L ) ∈ E , (u, L ) ∈ E represents i j i j In addition, for user u and tourism service item l in the the paired training data. E indicates that there is a record of test set, r represents the predicted score generated by the ul access records. E Q represents an unobserved visit record. recommendation algorithm, r represents the actual score ul y and y , respectively, represent the user’s preference for uL uL i j of user u on item i, and hence, the mean square error (MSE) the target POI L and L . δ(·) is the sigmoid function. Φ � i j can be defined as follows: c C φ, {ω } represents all model parameters that can be c�1 trained. λ represents the L regularization parameter that 1 MSE � r − r , (7) ul ul controls overfitting. -e small batch Adam optimization g�1 algorithm is used to optimize the prediction model and where G is the number of observed values in the test set. update the model parameters. Although deep learning models have strong representation capabilities, they often have the problem of overexpression 4.2. Performance Comparison with Comparison Algorithm [24]. Dropout is an effective solution to prevent neural net- works from overfitting. To make the model generalize well to 4.2.1. Influence Analysis of Model Characteristics. Since each unobserved data, Dropout is used in training. Dropout ran- feature in the recommendation model will have a certain impact on the recommendation results, user preferences domly deletes specific nodes and the neural network nodes of the information self-encoding layer with a certain probability. (UP), geographic factor (GF), and thematic factor (TF) are successively added to the proposed model, and experiments are carried out. -e comparison results under Pre@5, Pre@ 3.6. Generate Recommendation List. After the network 10, and Pre@20 are shown in Figure 3. model is trained using the above optimization algorithm, the As can be seen from Figure 3, the proposed model in- vector information of tourism service items and users can be tegrates user preference features, geographical factor fea- input into the network model. -rough the trained network tures, and topic factor features, and its recommendation model, the user’s prediction score of the tourism service item accuracy is better than the model with user preference and is obtained. -is value is used as a basis and arranged geographical factors. Taking Pre@5 as an example, its ac- according to size. Recommend the top N travel service items curacy is as high as 95%. At the same time, it can also be seen with higher scores to the user, generate a personalized that the recommendation effect obtained by the fusion of the recommendation list, and complete the recommendation. three-factor features is significantly better than that of the single factor features or the fusion of the two-factor features. 4. Experiment and Analysis -e data used in the experiment is from a tourism website, 4.2.2. Comparative Analysis of Accuracy and Recall. In and the basic information data example of its users is shown addition, by comparing the accuracy and recall of the in Table 1. -ere will be many comments in a scenic spot. proposed model with the models in reference [9, 11, 14] on 6 Journal of Robotics Table 1: Data examples of tourism service items. ID Name Label Commentary Huashan is the western mountain of the five mountains. It has a . . .“Huashan is really worth going. It is super i135728 Huashan unique landscape and is known as the most dangerous dangerous and the first mountain in the world” mountain in the world. . . . Lugu lake, with its natural and primitive folk customs and . . .“A beautiful and mysterious place, one of the i126239 Lugu lake beautiful natural scenery, is known as the “magical oriental holy lakes worth visiting in your life” . . . daughter country” Forbidden -e forbidden city is one of the largest and best preserved . . .“-e forbidden city after snow is really i135460 city wooden ancient buildings in the world beautiful. I Highly recommend it!” . . . the data set, the comparison results in the case of Pre@5, 1.00 Pre@10, and Pre@20 are shown in Figure 4, and the com- parison results in the case of Rec@5, Rec@10, and Rec@20 are shown in Figure 5. 0.80 It can be seen from Figures 4 and 5 that the accuracy and recall of the proposed model are significantly better than those of the other recommended models. Taking Pre@ 0.60 10 and Rec@10 as examples, their values are 88% and 83%, respectively, while those of the other models are less than 0.40 80%. -e proposed model uses the Gaussian kernel function to obtain the pairwise distance between the cor- responding POIs in the user check-in record, selects the 0.20 node2vec model to extract the network structure charac- teristics of the user’s social relationship, and captures the user’s preferences by the self-attention mechanism. Hence, 0.00 Pre@5 Pre@10 Pre@20 the overall recommendation effect is good. However, ref- UP erence [9] uses sequential pattern mining algorithm to UP+GF build the POI knowledge base and massive structured POI UP+GF+TF access sequence to realize the recommendation of the best tourism route, but there is no influence of geography and Figure 3: Comparison of @N results before characteristic factors of other factors. -erefore, the accuracy and recall rate of the the proposed model. recommendation scheme are low. Taking Pre@20 and Rec@ 20 as examples, both are less than 50%. Reference [11] formed a standardized travel data set by preprocessing the 1.00 data, such as word segmentation and denoising, stratified the scenic spots according to the location information of popular scenic spots, and recommended travel routes in 0.80 combination with the theme level and scenic spot char- acteristics, however, it did not deeply mine the users. 0.60 Hence, the performance of the recommendation model was poor. Reference [14] proposed a personalized travel rec- ommendation model based on weighted multi-information 0.40 constraint matrix decomposition scheme, which compre- hensively describes users and travel locations using photos, user access sequences, and text tags, and it allocates dif- 0.20 ferent weights in combination with the common access probability based on geographical distance, which can 0.00 achieve better travel route recommendation. However, Pre@5 Pre@10 Pre@20 because of the traditional method, the recommendation Proposed model Ref.[11] performance is lower than that of the proposed model using Ref.[15] Ref.[9] deep learning. Taking Rec@5 as an example, it is reduced Figure 4: Comparison of accuracy of pre @N data sets. by 9%. 4.2.3. MSE Analysis of Different Models. To demonstrate the It can be seen from Table 2 that the MSE values of recommended performance of the proposed model, it is reference [9, 11] are almost the same, only 0.031. Because of the lack of in-depth analysis of geography, user preferences, compared with reference [9, 11, 14]. -e results are shown in Table 2. and other factors, the recommended results deviate greatly Accuracy rate Accuracy rate Journal of Robotics 7 the user’s long-term preferences in each sequence, thereby 1.00 improving the accuracy of travel route recommendation. -e experimental results based on the pytoch deep learning framework show that the proposed model completes data 0.80 feature extraction and prediction using a deep learning net- work based on a self-attention mechanism, and it compre- hensively considers all kinds of data information. -erefore, its 0.60 Pre@10 and Rec@10 values are 88% and 83%, respectively, and the mean square error is 1.537, which has certain advantages in 0.40 the tourism route recommendation. At present, the extensive use of the knowledge map makes it possible to extract potential interactive represen- 0.20 tations that human beings cannot notice through this technology to make an effective recommendation has be- come a hot issue in research. Data often have diverse and 0.00 Rec@5 Rec@10 Rec@20 heterogeneous representations, such as the type of POI, the access time of POI, the traffic time of POI, the cost of POI, Proposed model Ref.[11] the location of POI, etc. Mining the attribute information of Ref.[15] Ref.[9] these entities using the knowledge map technology can make Figure 5: Comparison of accuracy of rec @N data sets. the recommendation system develop further. Table 2: MSE comparison results of different models. Data Availability Model MSE -e data used to support the findings of this study are in- Reference [9] 1.924 cluded within the article. Reference [11] 1.893 Reference [14] 1.705 Conflicts of Interest Proposed model 1.537 -e author declares that there are no conflicts of interest regarding the publication of this paper. from the actual route. Reference [14] uses the weighted multi-information constraint matrix decomposition method to realize personalized travel recommendation, which References considers many factors, however, it lacks a powerful learning [1] L. Cai, W. Wen, B. Wu, and X. Yang, “A coarse-to-fine user algorithm. -erefore, the accuracy of the recommendation preferences prediction method for point-of-interest recom- result is not high, and the MSE is 1.705. On the basis of mendation,” Neurocomputing, vol. 422, no. 3, pp. 1–11, 2021. preprocessing like data cleaning, the proposed model uses [2] Z. Zhang, C. Zou, R. Ding, and Z. Chen, “VCG: exploiting the deep learning algorithm to extract user features and carry visual contents and geographical influence for Point-of-In- out corresponding learning classification. It not only con- terest recommendation,” Neurocomputing, vol. 357, no. 9, siders the comment text information of users and tourism pp. 53–65, 2019. service items but also adopts the basic information of users [3] D. Yha, D. Bca, T. C. Jing, and Y. Zeng, “Privacy-preserving and tourism service items. -erefore, an ideal recommen- point-of-interest recommendation based on geographical and social influence,” Information Sciences, vol. 543, no. 8, dation scheme is obtained, and its MSE is only 1.537. In pp. 202–218, 2021. conclusion, the above results demonstrate the effectiveness [4] D. Yu, W. Wanyan, and D. Wang, “Leveraging contextual and superiority of the proposed recommendation model. influence and user preferences for point-of-interest recom- mendation,” Multimedia Tools and Applications, vol. 80, no. 8, 5. Conclusion pp. 1–15, 2021. [5] L. Chang, W. Chen, J. Huang, and C. Bin, “Exploiting multi- In recent years, with the popularization of the internet and the attention network with contextual influence for point-of- interest recommendation,” Applied Intelligence, vol. 51, no. 5, continuous development of information technology, people’s pp. 1–14, 2021. demand for tourism is richer and more diverse. Effective and [6] X. Xiong, S. Qiao, N. Han et al., “Where to go: an effective timely tourism service recommendation is of great significance point-of-interest recommendation framework for heteroge- to provide efficient and high-quality personalized tourism neous social networks,” Neurocomputing, vol. 373, no. 2, service recommendation. -erefore, based on the deep learning pp. 56–69, 2020. algorithm in the big data environment, a personalized tourism [7] Y. Si, F. Zhang, and W. Liu, “An adaptive point-of-interest route recommendation model that can be applied to the in- recommendation method for location-based social networks telligent service robot in the scenic hall is proposed. Aiming at based on user activity and spatial features,” Knowledge-Based the problems of long travel timespan of users and dynamic Systems, vol. 163, no. 1, pp. 267–282, 2019. changes of preferences, the proposed model uses the self-at- [8] S. Renjith, A. Sreekumar, and M. Jathavedan, “An extensive tention mechanism module to filter the POI features related to study on the evolution of context-aware personalized travel Recall rate 8 Journal of Robotics recommender systems,” Information Processing & Manage- ment, vol. 57, no. 1, pp. 102078.1–102078.19, 2020. [9] C. Bin, T. Gu, Y. Sun, and L. Chang, “A personalized POI route recommendation system based on heterogeneous tourism data and sequential pattern mining,” Multimedia Tools and Applications, vol. 78, no. 24, pp. 35135–35156, 2019. [10] G. Alexandridis, A. Chrysanthi, G. E. Tsekouras, and G. Caridakis, “Personalized and content adaptive cultural heritage path recommendation: an application to the Gournia and Çatalhoy ¨ uk ¨ archaeological sites,” User Modeling and User-Adapted Interaction, vol. 29, no. 1, pp. 201–238, 2019. [11] S. Du, H. Zhang, H. Xu, J. Yang, and O. Tu, “To make the travel healthier: a new tourism personalized route recom- mendation algorithm,” Journal of Ambient Intelligence and Humanized Computing, vol. 10, no. 9, pp. 3551–3562, 2019. [12] K. Song, M. Ji, S. Park, and I.-C. Moon, “Hierarchical context enabled recurrent neural network for recommendation,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 7, pp. 4983–4991, 2019. [13] L. R. Divyaa and N. Pervin, “Towards generating scalable personalized recommendations: integrating social trust, social bias, and geo-spatial clustering,” Decision Support Systems, vol. 122, no. 7, pp. 113066.1–113066.17, 2019. [14] D. Lyu, L. Chen, Z. Xu, and S. Yu, “Weighted multi-infor- mation constrained matrix factorization for personalized travel location recommendation based on geo-tagged photos,” Applied Intelligence, vol. 50, no. 1, pp. 1–15, 2020. [15] X. Cheng, “A travel route recommendation algorithm based on interest theme and distance matching,” EURASIP Journal on Applied Signal Processing, vol. 2021, no. 1, pp. 1–10, 2021. [16] L. A. Yan, A. Cyc, W. B. Ran, and C. S. Victor, “IMCRec: a multi-criteria framework for personalized point-of-interest recommendations,” Information Sciences, vol. 483, no. 7, pp. 294–312, 2019. [17] A. Exposito, ´ S. Mancini, J. Brito, and J. A. Moreno, “A fuzzy GRASP for the tourist trip design with clustered POIs,” Expert Systems with Applications, vol. 127, no. 8, pp. 210–227, 2019. [18] C. Villavicencio, S. Schiaffino, J. Andres Diaz-Pace, and A. Monteserin, “Group recommender systems: a multi-agent solution,” Knowledge-Based Systems, vol. 164, no. 1, pp. 436–458, 2019. [19] R. M. D’Addio, R. S. Marinho, and M. G. Manzato, “Com- bining different metadata views for better recommendation accuracy,” Information Systems, vol. 83, no. 7, pp. 1–12, 2019. [20] S. A. Yu, B. Hs, L. A. Chao, and L. Yin, “LSVP: a visual based deep neural direction learning model for point-of-interest recommendation on sparse check-in data,” Neurocomputing, vol. 446, pp. 204–210, 2021. [21] D. Yu, K. Xu, D. Wang, and T. Yu, “Point-of-Interest rec- ommendation based on user contextual behavior semantics,” International Journal of Software Engineering and Knowledge Engineering, vol. 29, no. 11, pp. 1781–1799, 2019. [22] Z. Huang, X. Lin, H. Liu, B. Zhang, Y. Chen, and Y. Tang, “Deep representation learning for location-based recom- mendation,” IEEE Transactions on Computational Social Systems, vol. 7, no. 3, pp. 648–658, 2020. [23] X. Wang and S. Kadolu, “Modeling uncertainty to improve personalized recommendations via Bayesian deep learning,” International Journal of Data Science and Analytics, vol. 2, no. 3, pp. 1–11, 2021. [24] X. Liu, C. Andris, and S. Rahimi, “Place niche and its regional variability: measuring spatial context patterns for points of interest with representation learning,” Computers, Environ- ment and Urban Systems, vol. 75, no. 5, pp. 146–160, 2019.
Journal
Journal of Robotics – Hindawi Publishing Corporation
Published: Jan 29, 2022