If we want to develop a context-aware recommender system in domains (such as movies, other products) which parameters will play the key role?. As far as my knowledge is concerned, i have noticed that most of the researchers has considered time, location, age etc. to recommend items to users while designing such systems. However, apart from these parameters, what other key parameters can play important role in this perspective?
Dear Zafar,
User emotions could be one the parameter when you want to design context-aware RS. Emotions can be obtained based on the user's latest post (i.e. FB status, tweets, etc). Text mining + NLP can be used to extract emotions. Also, the sentiment analysis is also a reliable technique to capture emotions.
You can refer to this article for further reading - Comparison and Evaluation of Recommendation Systems on Social Networks
Article Comparison and Evaluation of Recommendation Systems on Social Networks
Okay, think about a cross-domain recommendation. For instance, you integrate your context-aware RS to recommend places, with user Twitter account. When a user check-in at a certain location (i.e. London King's Cross tube station) and he posts a tweet (i.e. "What a lonely day :( ") on his Twitter, there are two different useful information.
(emoji also describes user's emotions, but cross checking between emojis and user tweets is essential)
Hence, we know that the user is currently at the KC tube station and today probably not a good day for him. If he tries to find a place to go, after that, the RS should be able to recommend a place nearby that will make him happy (i.e. a coffee shop - this is actually based on my experience, I normally go to a coffee shop if I am angry, sad).
Hope it helps.
Abdul Hadi Mohamad,
The information your provided are very useful and the example you provided is very awesome. Along with learning, i have read this answer with positive emotions as the answer is pretty clear and easy to understand. I think companion can be another context aware feature, in which we need to know about the companion of the person whom system is going to recommend items. For instance, in case of movies recommendation, if we are with kids then the system should recommend us cartoons and other funny movies instead of horror and romantic.
It is hard to know if you stay at home and there are kids around you, unless you tell the system there are kids with you, sit next to you Or if you choose to restrict the content while watching movie, it could probably one of the indicators.
However, if we are looking into a real-life application (i.e., Netflix), the application separates movies into a few distinct categories, and Kids movies is one of them. If you watch any kids movies for once or twice, the system will start recommend kids movies to your account. Your action of watching kids movies is actually telling the system there are kids associated with this account and hence, kids movies should be recommended and restrict all unnecessary contents (i.e., horror, romantic, etc).
Hope it helps :-)
Abdul Hadi Mohamad, do u have other parameters in your mind which could be considered in context-aware recommender systems?. Some people believe that for context-aware recommender systems we need smartphone and other devices which captures the current situation and context for a particular user. I want your opinion on this as well.
Dear Zafar,
Indeed, device is also a factor in which Context-Aware RS plays a vital role in. Besides how contextual information is captured, the type of device is also another thing. We tend to do different things on different device. For instance, I have 1 account associates with 3 different devices, including Ipad, Iphone, Macbook.
I usually play games on my Ipad, while on my iphone, I just use it for answering phone calls, texts and stuff, and I normally watch movies on my macbook (these are only examples). App recommender in this case, should consider the type of device used in the examples. Games should be recommended for Ipad, while any messaging app (such as Wechat, Whatsapp) should be recommended for my phone, and any movie app should be recommended for my macbook.
While context information should be treated differently with user preference, I think "device usage" or "device type" can also be one of the factor or input, whatever name you want to call it.
Other factors such as festive seasons, locations, emotions, and health are also useful.
- Badges
- Science topic