Introduction
-
Proposing a personalized POI route framework based on context and explicit demographic data using an asymmetric topic model.
-
Utilizing the term frequency inverse document frequency technique to calculate the similarity between contextual factors.
-
Representing a novel and improved CF method using the Markov function and user preferences to find the preferences.
Background knowledge
Approaches of topic model-based recommendation
Approaches of context-aware recommendation
Approaches of sequential patterns’ mining recommendation
Year | Ref. no. | Recommender system paradigms | ||||||||
---|---|---|---|---|---|---|---|---|---|---|
CF | CA | DB | Personalized | Sequence | Topic | Descriptions | Dataset | Advantages/potential limitations | ||
2014 | [27] | ✔ | ✔ | ✔ | A context-aware hierarchical Bayesian method Using topic regression with social matrix factorization | Epinions | Using spectral clustering for user-item Reduced complexity Not applicable for cold start No asymmetric similarity measured No personalized | |||
2015 | [45] | ✔ | ✔ | Used user preferences Employing contextual pre-filtering | Flickr | Improved accuracy Reduced complexity Small number of contextual parameters No asymmetric measure | ||||
2015 | [50] | ✔ | ✔ | ✔ | Using geo-tagged photos Considering user interest and sequences | Flickr | Considering probability No personalized recommendation No demographic information | |||
2015 | [2] | ✔ | ✔ | ✔ | Using the topics about user preference Using geo-tagged photos | Flickr | Dealing with data sparsity Consider user preference Not applicable for cold start | |||
2015 | [43] | ✔ | ✔ | ✔ | ✔ | Based on the topic distribution of his travel histories Considering season and weather as context information | Flickr | Using geo-tagged photos Utilizing a topic model to mine the preference of a user Not applicable for cold start and sparsity problems | ||
2016 | [38] | ✔ | ✔ | ✔ | ✔ | Using travelogues and community-contributed photos and the heterogeneous metadata (geo-location, and date taken) | Flickr and IgoUgo | Recommend a travel sequence Considering tags Not applicable for topic and cold start | ||
2017 | [40] | ✔ | ✔ | Context-aware probabilistic matrix factorization Modeling the social correlations Aggregated (LDA) model | Twitter and Foursquare | Modeling the topic model Using the textual, geographical, social, categorical, and popular information Dealing with data sparsity Not applicable for cold start | ||||
2017 | [51] | ✔ | ✔ | ✔ | ✔ | A hybrid context-aware recommender system with CF and GSP Personalized | LMS | Including context pre-filtering Alleviate cold start and sparsity problems | ||
2017 | [52] | ✔ | ✔ | ✔ | ✔ | ✔ | Location-based recommender system Preferred time-aware route planning | Gowalla and Foursquare | Considering the geographical, social, and temporal information of users Including user preferences Not personalized | |
2017 | [39] | ✔ | ✔ | ✔ | ✔ | Considering the user's interest and time frame Using Markov-topic | Flickr | Using clustering Employing the LDA Not applicable for cold start | ||
2018 | [53] | ✔ | ✔ | ✔ | Using sequence patterns Based on the semantic model Context-based (time and spatial data) | Flickr | Utilizing user preferences Based on geo-tagged photos No clustering Not applicable for cold start | |||
2018 | [18] | ✔ | ✔ | Personalized top-n sequential recommendation Using convolutional embedding recommendation | Gowalla Foursquare Tmall MovieLens | Dealing with data cold start Using user preferences and sequential patterns Not applicable for sparsity issue | ||||
2018 | [49] | ✔ | ✔ | ✔ | ✔ | A trip recommendation for tourists Considering the interests of users and sequences | Flickr | Personalized Not applicable for cold start No demographic information | ||
2018 | [54] | ✔ | ✔ | ✔ | ✔ | A personalized an itinerary recommendation with time constraints LBSN—exploiting geographical features and social relationships | Gowalla | User-based CF with time preference Considering the visiting time of locations Not applicable for cold start | ||
2019 | [55] | ✔ | ✔ | ✔ | Hybrid location-based travel recommender system Personalized travel recommendations | Trip advisor | Using swarm intelligence algorithms No asymmetric similarity measure | |||
2019 | [32] | ✔ | ✔ | ✔ | Personalized itinerary recommendation Integrates POI textual contents, historical user, and POI categories | Flickr | Context-based Alleviates the cold start No asymmetric similarity measure | |||
2019 | [19] | ✔ | ✔ | ✔ | A POI route recommender framework Using sequential pattern mining | Flickr | Dealing with data cold start and sparsity Generating different fine-grained candidate POI routes | |||
2019 | [56] | ✔ | ✔ | Personalized location recommendation Using Matrix factorization | Nokia mobile data | Using demographic features Personalized Dealing with data cold start | ||||
2020 | [4] | ✔ | ✔ | ✔ | A context-aware tourism Recommendation system Semantically clustered | Trip advisor | Using users' reviews on social networks to discover user's preferences Not applicable for cold start | |||
2020 | [57] | ✔ | ✔ | ✔ | Proposing a multi-level model Utilizing demographic information | Trip advisor | Using user preferences Dynamic contextual information Not applicable for cold start | |||
2020 | [1] | ✔ | ✔ | A tourist recommendation system Using trust criteria and contextual data | Trip advisor | Employing graph clustering Not applicable for cold start | ||||
2020 | [58] | ✔ | ✔ | Using an asymmetric measure Considering user factors | Movielens | Employing the probability density distribution Not applicable for cold start | ||||
2020 | [59] | ✔ | ✔ | Utilized the visual contents Using probabilistic Matrix factorization model | Flickr | Personalized Alleviates the cold start and data sparsity conditions No demographic information No asymmetric similarity measure | ||||
2020 | [13] | ✔ | ✔ | ✔ | ✔ | ✔ | Tourist recommendations based on demographic and context-aware Personalized | Flickr | Using asymmetric similarity measure Consider limited context parameters | |
2020 | [15] | ✔ | ✔ | ✔ | ✔ | Personalized travel recommendation based on geo-tagged photos Using matrix factorization | Flickr | Using contextual information, text information, and photo tags Not applicable for cold start | ||
2021 | [60] | ✔ | ✔ | ✔ | Personalized sequential pattern Considering decision tree Multi-label classification | CDNow-RFM and msdt2multi-valued | Using sequential pattern mining Deal with the cold start Not consider contextual data Not personalized | |||
2022 | [44] | ✔ | ✔ | ✔ | A personalized POI recommender system Using heterogeneous graphs | Trip advisor | Considering POI categories and periods No demographic information | |||
2022 | [24] | ✔ | ✔ | ✔ | ✔ | ✔ | A recommender system using CF and sequence patterns mining Based on geo-tagged photos | Flickr and STS | Using Demographic Data Using asymmetric similarity Deal with the cold start | |
TopicSeqHybrid | ✔ | ✔ | ✔ | ✔ | ✔ | ✔ | Our framework |
The proposed method
Problem identification
Offline phase
Data preprocessing
Enriching geo-tagged photo with contextual information
Time context | Visit day | Saturday, Sunday | Weekend |
Monday, …, Friday | Working day | ||
Visit time | 06:00–12:00 | Morning | |
12:00–18:00 | Afternoon | ||
18:00–06:00 | Night | ||
Visit season | March, April, and May | Spring | |
June, July, and August | Summer | ||
September, October, and November | Fall | ||
December, January, and February | Winter | ||
Weather context | Temperature | > 34 °C | Hot |
18–34 °C | Warm | ||
< 18 °C | Cold | ||
Weather | Sunny, clear sky | Sunny | |
Cloudy, broken clouds, scattered clouds | Cloudy | ||
Rain, fog | Rainy | ||
Snow, snowfall | Snowy |
Finding POIs
Producing the profile of points of interests (POIs)
User–POI detection
Topic-based calculation of user–user asymmetric schema
Sequence extraction
SPM algorithm
Online phase
Enriching user queries by contextual data
Pre-filtering based on context
Combination of the recommendations
Recommendation
Candidate trip pattern stage
Simulations and experimental evaluation
Evaluation dataset
ID | Owner | Title | Time-taken | Tags | Latitude | Longitude |
---|---|---|---|---|---|---|
68923891584 | 5413236@N04 | British museum | 9/15/2018 16:41 | British museum, London, sculpture | 51.51945 | − 0.12606 |
Images | Tourists | Location | |
---|---|---|---|
Raw | Filtered | Filtered | |
49,999 | 44,263 | 456 | 2957 |
The evaluation metrics
Comparison approaches
Methods | #References | Description |
---|---|---|
(CF) | [6] | Cosine similarity |
(PR) | [50] | Public popularity |
(Pre_CA-CF) | [73] | Contextual pre-filtering similarity measure |
(CA-CF) | [74] | Jaccard measure |
(ACA-CF) | [7] | Cosine similarity + asymmetric schema with Jaccard measure |
(GSP-CACF) | [51] | CF and GSP |
(CA-MSDT) | [60] | CA and decision tree classification |
(Prefix-CSTR) | [50] | Prefix-span algorithm |
(ADBCACF) | [13] | Asymmetric CF and demographic data |
(SeqHybrid) | [24] | Our previous work describes a sequential recommender system that combines context-awareness, demographic-based, and asymmetric CF |
Experimental results
Impact of parameter β
The effect of the neighborhood numbers
The impact of the highest suggestions
Evaluation of TopicSeqHybrid with the Gowalla
Evaluation of TopicSeqHybrid by NDCG metric
Example trips
City | Photos | Tourists | POI visits | Trip seq. |
---|---|---|---|---|
Toronto | 1,57,500 | 1390 | 39,410 | 6053 |