[GSoC 2013] Proposal by Pranjal Goswami
Clone this wiki locally
Name : Pranjal Goswami
University: Indian Institute of Technology, Kharagpur, India.
ThinkUp provides extensively analyzed data and portrays them to the user in a unique fashion. In the limited time I had to look into the project, I have observed that the application extensively harnesses the abilities of the twitter API. The insights cater interesting facts about their actions on the social network.
Work on present insights and some new insights:
Currently the webapp performs well for twitter. I will enhance the performance of the Facebook plug-in as a lot of users are inactive on twitter but relatively very active on Facebook. The data comparison of user’s activity on different social networks can be an additional insight.
Apart from the algorithms used to extract these insights, the visualization of the data is equally important. Currently the insights show additional data using the Google Chart API. I will show more data in form of graphs using d3 as it allows a control of the visuals.
Finding people with similar interests
This feature can identify followers with similar interests and can suggest the user to follow them back. The 'similar interests' are obtained by performing topical clustering over documents (here tweets, posts) and obtaining labeled clusters. Standard algorithms like K-means(http://en.wikipedia.org/wiki/K-means_clustering) or other more tunable algorithms offering better semantic matching (e.g. phrase matching) like Lingo (http://project.carrot2.org/publications/osinski-2003-lingo.pdf) could be used. These cluster labels for each follower can then be matched against the user's clusters and a similarity score could be provided (as simple as number of label keywords matching). Followers with similarity above a certain threshold would be suggested to follow back.
Use of this feature: Following people with common interests has always been one of the prime reason why people use Twitter. This feature offers the 'follow' suggestions from user's own followers.
Using Google’s sentiment analysis
- To get a response/feedback about a place wants to visit(restaurant/hotel/holiday spot), the application will perform a sentiment analysis on the tweets with #tag of the keyword(place) and then returns a feedback/rating/score for the particular place.
- For every new tweet of the user, the hash-tags/keywords used will be grabbed and a sentiment analysis will be performed on the tweets of followers/posts by friends containing the similar keywords/hash-tags. The result will show how similar is the user’s response with his followers/friends. The keywords will primarily be proper nouns. For Example a user tweets: @pranjal-goswami : today chirs #gayle was awesome!!!!
The application will show the similarity of this tweet with the responses of the followers who tweeted about gayle, using the sentiment analysis
- Using Sentiment Analysis on the posts and tweets of the user, the mood changes of the user can be gauged. Also, by making a sensitivity analysis of the replies, we can gauge the overall feedback to that tweet or post. “Your tweet _____ received a negative response”.
Usually, it happens that a person visits a place and tells about it to other people. Many a times, people check-in at the place or post it as a status message. They may also recommend this to their friends. This insight gives an idea of the impact of this positive feedback. Implementing this feature is easy as check-in data is available using facebook graph API
Apart from upcoming milestones, the user can be notified when a particular milestone is reached. For example a photo reaches more than the present maximum number of likes. Similar is for notes, videos and posts.
Fetch Music and Photographs:
The posts and tweets can be used to grab the interest of the user. Also, the ‘interests’ from facebook API can be grabbed and stored to fetch related pictures from flickr and music from youtube adding flavor to the present application.
Generating breadcrumb trail
Using the geoLocation data from check-ins using different social network applications, a breadcrumb trail of the user can be visualized using the google map API. This data can be generated weekly and then comparison can be shown as an insight. “You have checked-in at a lot more places than last week”