Rich geo-textual data is available online and the data keeps increasing at a high speed. We propose two user behavior models to learn several types of user preferences from geo-textual data, and a prototype system on top of the user preference models for mining and search of geo-textual data to support personalized maps. Different from existing recommender systems and data analytics systems, our technology highly personalizes user experience on maps and supports several applications, including user mobility & interests mining, opinion mining in regions, user recommendation, point-of-interest (POI) recommendation, and querying and subscribing of geo-textual data.
Offline components: Our technology contains three offline components, namely Data Collector, User Behavior Models, and Indexes.
1) Data Collector: The data collector grabs and integrates the historical data from difference sources (e.g., Twitter, Foursquare and Yelp). It also keeps monitoring the streaming data to support pushing news for the online subscription component.
2) User Behaviour Models: The user behaviour models analyze user’s historical geo-textual posts to reveal the user’s interests in regions, topics, categories, and characteristics. The user preferences are then used by all the online applications.
3) Indexes: To efficiently support updating and querying of geotextual data (Foursquare check-ins and Yelp reviews), we use a simple hybrid data structure that combines a shallow quadtree and inverted file to index geo-textual data, which is called IQ-tree. The IQ-tree is also used to index subscription queries if there are a large number of such queries.
Online components: The online components include solutions for different applications using the user preferences learnt by the user behaviour models in the offline part. The applications include mobility & interests mining, aspect analysis of regions, user recommendation to business, personalized POI recommendation and personalized querying & subscription.
a) Location recommendation apps – incorporate the user’s mobility pattern to provide recommendations close to the user’s mobility and interests. As the user’s interests are well-profiled, the recommendation will be conducted according to the matches between the user’s interests and the location’s attributes.
b) Business data analytics apps – as the API can analyze the characteristics (e.g., price, environment) described in the reviews, it can summarize the overall satisfaction of a particular region to help evaluate the business.
Recommender systems has been a hot topic in the market. Many e-commercial companies and location-based service providers (e.g., Amazon, Foursquare, Yelp) provide recommendations on items, restaurants, etc. Moving from the naïve methods to the modern technologies, recommender systems are getting more and more intelligent and requires lesser and lesser explicit user input. Apart from recommendations, the user’s feedback can be valuable information to explore with regard to business decision support.
To the best of our knowledge, most of the existing location recommender systems such as Foursquare and Yelp still require users to input location and category to narrow down the recommendation list. Also, they cannot provide explanation of why a location is recommended, which is regarded as an important feature for recommender systems in recent years. Our technology fills the gap of the existing systems by automatically analyzing the reviews from the users and then generating the profiles of the users from both angles of topical interests and spatial mobility. The new features provided by our technology has potential market as a unified solution for both intelligent recommendation and business analytics.
The system benefits two kinds of end users – (1) consumer and (2) business owners. For consumers, they can receive recommendations that are personalized to them even if they do not specify any explicit query such as “restaurant”, as the system can automatically learn about the consumers’ interest. For business owners, the system can automatically analyse the reviews written by consumers and thus provide business-level analysis. For example, recommending advertising to users particular highlights (e.g., discounts). For another example, aspect satisfaction analysis help business owners to understand which characteristics are good/bad in different regions.