Natwar Modani

Summarizing Multimedia Content

Today multimedia content comprising both text and images is growing at a rapid pace. There has been a body of work to summarize text content, but to the best of our knowledge, no method has been developed to summarize multimedia content. We propose two methods for summarizing multimedia content. Our novel approach explicitly recognizes two desirable, normative characteristics of a summary - good coverage and diversity of the respective text and images, and that text and images should be coherent with each other. Two methods are examined - graph based and a modification to the submodular approach. Moreover, we propose a metric to measure the quality of a multimedia summary which captures coverage and diversity of text and images as well as coherence between the text and images in the summary. We experimentally demonstrate that the proposed methods achieve good quality multimedia summaries.

Creating Diverse Product Review Summaries: A Graph Approach

Product reviews play an influential role for the e-commerce websites, as consumers leverage them during the purchase decision process. However, the volume of such reviews can be overwhelming for a web user to comprehend the gist of overall information communicated by other consumers. In this paper, we address the problem of summarizing user contributed product reviews, having certain properties that differentiate them significantly from summarizing of traditional text articles. We propose suitable summarization algorithms that capture useful information with minimum redundancy and maximum information. We present a graph based formulation using a fast and scalable greedy algorithm for the review summarization problem. Our approach provides a rich model that makes certain sentences more rewarding based on their properties, in addition to their relation to the other reviews. We evaluate and show that our proposed algorithm outperforms other state-of-the-art summarization algorithms with significance level of 0.01 using automatic evaluation.

Fairness Aware Recommendations on Behance

Traditionally, recommender systems strive to maximize the user acceptance of the recommendations, while more recently, diversity and serendipity have also been addressed. In two-sided platforms, the users can have two personas, consumers who would like relevant and diverse recommendations, and creators who would like to receive exposure for their creations. If the new creators do not get adequate exposure, they tend to leave the platform, and consequently, less content is generated, resulting in lower consumer satisfaction. We propose a re-ranking strategy that can be applied to the scored recommendation lists to improve exposure distribution across the creators (thereby improving the fairness), without unduly affecting the relevance of recommendations provided to the consumers. We also propose a different notion of diversity, which we call representative diversity, as opposed to dissimilarity based diversity, that captures level of interest of the consumer in different categories. We show that our method results in recommendations that have much higher level of fairness and representative diversity compared to the state-of-art recommendation strategies, without compromising the relevance score too much. Interestingly, higher diversity and fairness leads to increased user acceptance rate of the recommendations.

Generating Multiple Diverse Summaries

Authors often re-purpose existing content to create shorter versions for other channels. Automatic summarization techniques can be used to generate a candidate content that can be further fine-tuned by the author. Existing work in automatic summarization primarily focus on providing a single succinct summary. However, this may not suit the needs of a content author or curator, who may want to repurpose/select the content from several alternative candidates. In this paper, we propose an approach to generate multiple diverse summaries, so that authors can choose an appropriate summary without compromising on the summary quality. Our approach can be utilized in conjunction with a large class of extractive summarization techniques, and we illustrate our approach with several summarization techniques. We experimentally show that our approach results in fairly diverse summaries, without compromising the quality of the summaries with respect to the single summary generated by the corresponding base methods.