Geo-Teaser: Geo-Temporal Sequential Embedding Rank for POI Recommendation

Zhao, Shenglin; Lyu, Michael R.; King, Irwin

doi:10.1007/978-981-13-1349-3_4

Cited by 67 publications

(129 citation statements)

References 37 publications

Supporting

Mentioning

127

Contrasting

Order By: Relevance

“…Three temporal attributes of periodicity, consecutiveness, and non-uniformness [8,10,22,85,93,18] have already been used in location recommendation systems. Periodicity studies the recurrent trips in users' mobility patterns (e.g.…”

Section: Motivationmentioning

confidence: 99%

“…Previous studies have solely considered one or two temporal scales such as hour [79,22,85,89] or day, and weekday/weekend periods [93,22] to circumvent overfitting concerns [89]. In addition, we model the recency attribute, which is two-fold: Firstly, users with recent check-in activity must obtain higher similarity weights in Collaborative Filtering (CF).…”

Section: Motivationmentioning

confidence: 99%

“…Previous work [93,79,18,89,15,22,85,87] has integrated a single or two temporal effects such as the hour of the day or day of the week cycles into POI recommendation models. Therefore, we initially exploit a uni-aspect time-related influence in the location recommendation.…”

Section: Location Recommendation Using Lbsn Datamentioning

confidence: 99%

“…With respect to time, POI recommenders have so far employed the three temporal attributes of periodicity, consecutiveness, and non-uniformness [8,10,22,85,93,18]. Periodicity [10] states that a user's movements at different locations has an approximate periodical replication.…”

Section: Multi-aspect Temporal Influence (Mati) (Chapter 4)mentioning

confidence: 99%

“…a sample supermarket chain in Australia closes at 6pm except on Thursdays when they serve the customers until 9pm), hence discrete-time can also be considered when studying the role of multi-aspect temporal influence in location recommendation systems. However, previous work [93,79,18,89,15,22,85,87] that integrated discretized temporal information such as the hour of the day or day of the week cycles into POI recommendations considered only single or two temporal granularities to avoid complexity and overfitting issues [89]. In reality, time is multi-aspect and user check-in behaviors are simultaneously influenced by multiple temporal effects with different cycles.…”

Section: Multi-aspect Temporal Influence (Mati) (Chapter 4)mentioning

confidence: 99%

See 4 more Smart Citations

Location Inference and Recommendation in Social Networks

Hosseini¹

View full text Add to dashboard Cite

With the ubiquity of GPS-enabled smartphones, Location-Based Service (LBS) as a prominent product of social networks has become an essential part of our daily lives. People can easily socialize and share their check-in data (location, time, text, photo and etc.) via Location-Based Social Networks (LBSN). Through mining the check-in dataset, Point-Of-Interest(POI) recommendation systems assist users in exploring new attractive venues.The primary challenge regarding POI recommendation is to suggest a list of new interesting locations to the query user. Specifically, the excessive sparsity observed in user-location matrices is the main problem. The well known Collaborative Filtering (CF) methods are commonly used with other factors like geographical, social, context-oriented (e.g. text contents) and temporal effects to promote the effectiveness of the location recommendation systems. Despite the vital role of temporal influence, an insufficient amount of research has been devoted to considering the time factor in location recommendation. While we have an insufficient number of records regarding a user's check-in at a particular location, predicting the time of the visit seems more problematic. We have dedicated part of our research to study various aspects of the temporal influence in the recommendation.Additionally, we use Twitter textual contents to extract a set of spatial phrases associated with each region. Such an act enriches the textual contents of the local POIs. This process enhances location recommendation systems, as it facilitates textual similarity among POI tags and the tweet history of the query user.In short, we aim to address the problems involved with both aspects of Location Inference and Recommendation in social networks. Our research in this thesis has four parts. Firstly, we define a problem which merely considers a single temporal aspect to enhance the performance of a location recommendation task. Subsequently, we develop a probabilistic model which detects a user's temporal orientation based on visibility weights of POIs visited by her during weekday/weekend cycles. While this method is limited to a single temporal scale, the idea can be adapted to other time-related periodic cycles (e.g. daily home-work return trips). Secondly, we argue that the majority of existing methods merely concentrate on a limited number of temporal scales and neglect others. We propose a probabilistic generative model, named after Multi-aspect Time-related Influence (MATI) which employs the user's check-in log to detect her temporal mobility pattern under various scales (e.g. minute, hour, day and so on). It then performs recommendation using multi-aspect temporal correlations between the query user and proposed locations. Thirdly, we further study the role of the time factor in recommendation models. We define a new problem to jointly model a pair of heterogeneous time-related 2 effects (recency and the subset feature) in location recommendation. To address the challenges, we propose a generative model ...

show abstract

Section: Motivationmentioning

confidence: 99%

Section: Motivationmentioning

confidence: 99%

Section: Location Recommendation Using Lbsn Datamentioning

confidence: 99%

Section: Multi-aspect Temporal Influence (Mati) (Chapter 4)mentioning

confidence: 99%

Section: Multi-aspect Temporal Influence (Mati) (Chapter 4)mentioning

confidence: 99%

See 3 more Smart Citations

Location Inference and Recommendation in Social Networks

Hosseini¹

View full text Add to dashboard Cite

show abstract

Learning Cheap and Novel Flight Itineraries

Karamshuk¹,

Matthews²

2019

Machine Learning and Knowledge Discovery in Databases

View full text Add to dashboard Cite

We consider the problem of efficiently constructing cheap and novel round trip flight itineraries by combining legs from different airlines. We analyse the factors that contribute towards the price of such itineraries and find that many result from the combination of just 30% of airlines and that the closer the departure of such itineraries is to the user's search date the more likely they are to be cheaper than the tickets from one airline. We use these insights to formulate the problem as a trade-off between the recall of cheap itinerary constructions and the costs associated with building them. We propose a supervised learning solution with location embeddings which achieves an AUC=80.48, a substantial improvement over simpler baselines. We discuss various practical considerations for dealing with the staleness and the stability of the model and present the design of the machine learning pipeline. Finally, we present an analysis of the model's performance in production and its impact on Skyscanner's users.

show abstract

Comparison of Sentiment Analysis and User Ratings in Venue Recommendation

Wang

Ounis

Macdonald

2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Venue recommendation aims to provide users with venues to visit, taking into account historical visits to venues. Many venue recommendation approaches make use of the provided users' ratings to elicit the users' preferences on the venues when making recommendations. In fact, many also consider the users' ratings as the ground truth for assessing their recommendation performance. However, users are often reported to exhibit inconsistent rating behaviour, leading to less accurate preferences information being collected for the recommendation task. To alleviate this problem, we consider instead the use of the sentiment information collected from comments posted by the users on the venues as a surrogate to the users' ratings. We experiment with various sentiment analysis classifiers, including the recent neural networks-based sentiment analysers, to examine the effectiveness of replacing users' ratings with sentiment information. We integrate the sentiment information into the widely used matrix factorization and GeoSoCa multi feature-based venue recommendation models, thereby replacing the users' ratings with the obtained sentiment scores. Our results, using three Yelp Challenge-based datasets, show that it is indeed possible to effectively replace users' ratings with sentiment scores when state-of-the-art sentiment classifiers are used. Our findings show that the sentiment scores can provide accurate user preferences information, thereby increasing the prediction accuracy. In addition, our results suggest that a simple binary rating with 'like' and 'dislike' is a sufficient substitute of the current used multi-rating scales for venue recommendation in location-based social networks.

show abstract

Geo-Teaser: Geo-Temporal Sequential Embedding Rank for POI Recommendation

Cited by 67 publications

References 37 publications

Location Inference and Recommendation in Social Networks

Location Inference and Recommendation in Social Networks

Learning Cheap and Novel Flight Itineraries

Comparison of Sentiment Analysis and User Ratings in Venue Recommendation

Contact Info

Product

Resources

About