KDE04-Location Inference for Non-geotagged Tweets in User Timelines


As social media clients are increasingly going mobile, different location based Services (LBS) have been sent via web-based social media like Twitter. The success of them heavily relies upon the availability and accuracy of clients’ location data. Be that as it may, just a little fraction of tweets are geo-tagged. In this way, it is important to infer locations for tweets with a specific end goal to attain the purpose of LBS. In this project, we handle this issue by scrutinizing Twitter client timelines. At first, we split every client’s tweet timeline temporally into a number of clusters, each tending to imply a distinct location. In this manner, we adjust two machine learning models and outline classifiers that order each tweet cluster into one of the pre-defined location classes at the city level. The Bayes construct show centers in light of the information gain of words with area suggestions in the client created substance. The convolutional LSTM show treats client created substance and their related areas as arrangements and utilizes bidirectional LSTM and convolution task to deduce areas. The two models are assessed on an expansive large real data set. The results propose that our models are effective at infferring locations  for non-geo tagged tweets and the models outperform thestate-of-the-art and alternative approaches significantly in terms of inference accuracy.

Leave a Reply

Your email address will not be published. Required fields are marked *