Real-time Analytics for Resource Mobilization in Engineering Diversity Hashtag Campaigns

Abstract

To counter the lack of diversity within the engineering profession and to promote equity, many activists utilize social media platforms to raise awareness of the issue. One of the goals of these activism campaigns is to break down stereotypes about who is an engineer and to make the diverse participation visible within the profession. Although prior work indicates that social media hashtag campaigns help promote diversity, it also outlines several challenges that need to be addressed for efficiency and increased inclusiveness. In particular, more work is needed to understand the initial phases of these campaigns because during that stage, efficient resource mobilization is needed to attract participants and make the campaign a success. For resource mobilization, campaigns need to identify and attract influential users and create a message that resonates with participants in real-time. By sharing effective messages, influential users can thus create momentum for the campaign. This dissertation research develops two analytical frameworks to assist in finding out (a) which types of users can be more influential for educating and informing people in real-time, and (b) what kind of messages are deemed more attractive and inspirational to the targeted audience in real-time. To develop the frameworks and to test them, data from two Twitter hashtag campaigns is used: #ILookLikeAnEngineer (19492 tweets over the course of two months) and #WomenInEngineering (8225 tweets over the course of 16 months). The first part of this dissertation addresses the problem of identifying user types in these hashtag campaigns in real-time. Our preliminary work showed that organizations and individuals both played a significant role in promoting these campaigns goals but in different ways, both quantitatively and qualitatively. Therefore, it was important to distinguish different kinds of users with this classification in mind. This problem was solved by advancing an analytical framework to classify users into individuals (male or female) or organizations in real-time. This framework uses the multimodality of the information available in one tweet from a given user, such as user’s profile picture, user’s name, user’s network metadata, linguistic and psychological characteristics of the tweet, and determines the user type of the given tweet. The proposed framework with real-time features outperformed other baselines with more than 6.65% accuracy increase for ILookLikeAnEngineer and showed that organizations were one of the influential users after female participants. The second part of this dissertation addresses the problem of identifying messages that resonate with users. Based on prior work in the literature, it was hypothesized that some latent features, i.e.: the cluster of various co-occurring hashtags along with relevant topics and relatable sentiments in the messages can be deciding factors to attract like-minded users. An analytical framework was developed that utilizes these features to predict if a given tweet will be retweeted in real-time. The proposed framework works with real-time data and outperformed other baselines with 6.61% and 8.25% accuracy increase respectively for ILookLikeAnEngineer and WomenInEngineering.