SociaDictor (Social Predictor) Overview

SociaDictor is an ongoing project at UC Riverside. Our goal is to bring the power of social and news big data to the user’s fingertips through a suite of Web tools.

We are continuously monitoring and analyzing data from social networks, forums and news feeds, and we provide intuitive visualization tools to interactively explore various dimensions of this massive information. 

SociaDictor Architecture

SocialPredictor ( 

The Keyword Trends tool offers a comprehensive and multi-dimensional picture of keyword-defined topics over time, and predicts their future trends. For example, one may want to view how the sentiment around the word "diabetes" is changing over time. Further, we predict how various features, like sentiment, are expected to change in the near future.

When inputting multiple keywords, Keyword Trends generates holistic feature predictions, that is, it considers all the keywords to predict the features of each one of them. For example, the prediction of the number of tweets on "diabetes" will be different if we submit "diabetes, health insurance" vs. just "diabetes". Intuitively, if the user believes that there is salient relationship between "diabetes" and "health insurance" mentions in the news or social networks, she can provide both terms to potentially generate more accurate predictions for the "diabetes" features.

The Data Prediction tool leverages the same historical social and news data as Keyword Trends to predict the future trends of a user-input time series, given a list of keywords that the user deems relevant to the time series. 

A user inputs a comma-delimited file that contains a list of (date, number) pairs (e.g., past monthly product sales, home prices in NYC, stock price of MCD) and a few relevant keywords (e.g., a product's name, "New York" or "$MCD"). Data Predictor combines the past values of the input time series, and the historical trends of the provided keywords in the indexed social networks and news articles, to generate predictions for the next few time periods (e.g., next two months or next two days). 

The key idea, as in Keyword Trends, is to give to the user the power to specify the keywords that she thinks are related to the input time series. For example, a user may want to predict the sales of the popular diabetes drug "Lantus". Then, she can generate a comma delimited file with past monthly sales, in the form of (date, sale amount) pairs, and input it to Data Prediction along with keywords "Lantus, diabetes". This intuitively means that the user believes that there is a salient relationship between the sales of this drug and the chatter in social and news networks about "Lantus" and "diabetes". Note that we do not just consider the number of tweets or news as possibly correlated features, but a wide range of features including the sentiment fluctuation for the keyword-related posts or articles.

Twitter Search ( 

A tool to search and view on the map tweets organized by their popularity or geolocation. This is mainly an auxiliary tool invoked from SocialPredictor.



Not all keywords are indexed. We provide an autocomplete functionality to easily view which keywords are available. We have focused on a few domains like healthcare, financials, locations, food and celebrities for now.


Contact Us

Your feedback or suggestions are always welcome. Please email us here.


Professor Vagelis Hristidis

Michael Brevard, Undergraduate

Shiwen Cheng, PhD Candidate

Moloud Shahbazi, PhD Candidate

Past Contributors:

William Ibekwe, BS, Alumnus

Abhijith Kashyap, PhD, Alumnus

Eduardo Ruiz, PhD, Alumnus

Jehan Sethna, MS, Alumnus