Manish Singh

Anson Bastos, Abhishek Nadgeri, Kuldeep Singh et al.

Learning on evolving(dynamic) graphs has caught the attention of researchers as static methods exhibit limited performance in this setting. The existing methods for dynamic graphs learn spatial features by local neighborhood aggregation, which essentially only captures the low pass signals and local interactions. In this work, we go beyond current approaches to incorporate global features for effectively learning representations of a dynamically evolving graph. We propose to do so by capturing the spectrum of the dynamic graph. Since static methods to learn the graph spectrum would not consider the history of the evolution of the spectrum as the graph evolves with time, we propose a novel approach to learn the graph wavelets to capture this evolving spectra. Further, we propose a framework that integrates the dynamically captured spectra in the form of these learnable wavelets into spatial features for incorporating local and global interactions. Experiments on eight standard datasets show that our method significantly outperforms related methods on various tasks for dynamic graphs.

Anson Bastos, Kuldeep Singh, Abhishek Nadgeri et al.

In this paper we present a novel method, $\textit{Knowledge Persistence}$ ($\mathcal{KP}$), for faster evaluation of Knowledge Graph (KG) completion approaches. Current ranking-based evaluation is quadratic in the size of the KG, leading to long evaluation times and consequently a high carbon footprint. $\mathcal{KP}$ addresses this by representing the topology of the KG completion methods through the lens of topological data analysis, concretely using persistent homology. The characteristics of persistent homology allow $\mathcal{KP}$ to evaluate the quality of the KG completion looking only at a fraction of the data. Experimental results on standard datasets show that the proposed metric is highly correlated with ranking metrics (Hits@N, MR, MRR). Performance evaluation shows that $\mathcal{KP}$ is computationally efficient: In some cases, the evaluation time (validation+test) of a KG completion method has been reduced from 18 hours (using Hits@10) to 27 seconds (using $\mathcal{KP}$), and on average (across methods & data) reduces the evaluation time (validation+test) by $\approx$ $\textbf{99.96}\%$.

Aditya Kusupati, Manish Singh, Kush Bhatia et al.

This paper develops the FastRNN and FastGRNN algorithms to address the twin RNN limitations of inaccurate training and inefficient prediction. Previous approaches have improved accuracy at the expense of prediction costs making them infeasible for resource-constrained and real-time applications. Unitary RNNs have increased accuracy somewhat by restricting the range of the state transition matrix's singular values but have also increased the model size as they require a larger number of hidden units to make up for the loss in expressive power. Gated RNNs have obtained state-of-the-art accuracies by adding extra parameters thereby resulting in even larger models. FastRNN addresses these limitations by adding a residual connection that does not constrain the range of the singular values explicitly and has only two extra scalar parameters. FastGRNN then extends the residual connection to a gate by reusing the RNN matrices to match state-of-the-art gated RNN accuracies but with a 2-4x smaller model. Enforcing FastGRNN's matrices to be low-rank, sparse and quantized resulted in accurate models that could be up to 35x smaller than leading gated and unitary RNNs. This allowed FastGRNN to accurately recognize the "Hey Cortana" wakeword with a 1 KB model and to be deployed on severely resource-constrained IoT microcontrollers too tiny to store other RNN models. FastGRNN's code is available at https://github.com/Microsoft/EdgeML/.

Anson Bastos, Kuldeep Singh, Abhishek Nadgeri et al.

We present the Evolving Graph Fourier Transform (EFT), the first invertible spectral transform that captures evolving representations on temporal graphs. We motivate our work by the inadequacy of existing methods for capturing the evolving graph spectra, which are also computationally expensive due to the temporal aspect along with the graph vertex domain. We view the problem as an optimization over the Laplacian of the continuous time dynamic graph. Additionally, we propose pseudo-spectrum relaxations that decompose the transformation process, making it highly computationally efficient. The EFT method adeptly captures the evolving graph's structural and positional properties, making it effective for downstream tasks on evolving graphs. Hence, as a reference implementation, we develop a simple neural model induced with EFT for capturing evolving graph spectra. We empirically validate our theoretical findings on a number of large-scale and standard temporal graph benchmarks and demonstrate that our model achieves state-of-the-art performance.

Manish Singh, Arpita Dayama

This work evaluates the effectiveness of spatiotemporal Graph Neural Networks (GNNs) for multi-store retail sales forecasting and compares their performance against ARIMA, LSTM, and XGBoost baselines. Using weekly sales data from 45 Walmart stores, we construct a relational forecasting framework that models inter-store dependencies through a learned adaptive graph. The proposed STGNN predicts log-differenced sales and reconstructs final values through a residual path, enabling stable training and improved generalisation. Experiments show that STGNN achieves the lowest overall forecasting error, outperforming all baselines in Normalised Total Absolute Error, P90 MAPE, and variance of MAPE across stores. Analysis of the learned adjacency matrix reveals meaningful functional store clusters and high-influence nodes that emerge without geographic metadata. These results demonstrate that relational structure significantly improves forecast quality in interconnected retail environments and establishes STGNNs as a robust modelling choice for multi-store demand prediction.

Nagendra Kumar, Gopi Ande, J. Shirish Kumar et al.

A large amount of content is generated everyday in social media. One of the main goals of content creators is to spread their information to a large audience. There are many factors that affect information spread, such as posting time, location, type of information, number of social connections, etc. In this paper, we look at the problem of finding the best posting time(s) to get high content visibility. The posting time is derived taking other factors into account, such as location, type of information, etc. In this paper, we do our analysis over Facebook pages. We propose six posting schedules that can be used for individual pages or group of pages with similar audience reaction profile. We perform our experiment on a Facebook pages dataset containing 0.3 million posts, 10 million audience reactions. Our best posting schedule can lead to seven times more number of audience reactions compared to the average number of audience reactions that users would get without following any optimized posting schedule. We also present some interesting audience reaction patterns that we obtained through daily, weekly and monthly audience reaction analysis.

Nagendra Kumar, Yash Chandarana, K. Anand et al.

Nowadays online social networks are used extensively for personal and commercial purposes. This widespread popularity makes them an ideal platform for advertisements. Social media can be used for both direct and word-of-mouth (WoM) marketing. Although WoM marketing is considered more effective and it requires less advertisement cost, it is currently being under-utilized. To do WoM marketing, we need to identify a set of people who can use their authoritative position in social network to promote a given product. In this paper, we show how to do WoM marketing in Facebook group, which is a question answer type of social network. We also present concept of reinforced WoM marketing, where multiple authorities can together promote a product to increase the effectiveness of marketing. We perform our experiments on Facebook group dataset consisting of 0.3 million messages and 10 million user reactions.

Nagendra Kumar, Rakshita Nagalla, Tanya Marwah et al.

Social media is currently one of the most important means of news communication. Since people are consuming a large fraction of their daily news through social media, most of the traditional news channels are using social media to catch the attention of users. Each news channel has its own strategies to attract more users. In this paper, we analyze how the news channels use sentiment to garner users' attention in social media. We compare the sentiment of social media news posts of television, radio and print media, to show the differences in the ways these channels cover the news. We also analyze users' reactions and opinion sentiment on news posts with different sentiments. We perform our experiments on a dataset extracted from Facebook Pages of five popular news channels. Our dataset contains 0.15 million news posts and 1.13 billion users reactions. The results of our experiments show that the sentiment of user opinion has a strong correlation with the sentiment of the news post and the type of information source. Our study also illustrates the differences among the social media news channels of different types of news sources.

Tapan Sahni, Chinmay Chandak, Naveen Reddy Chedeti et al.

As microblogging services like Twitter are becoming more and more influential in today's globalised world, its facets like sentiment analysis are being extensively studied. We are no longer constrained by our own opinion. Others opinions and sentiments play a huge role in shaping our perspective. In this paper, we build on previous works on Twitter sentiment analysis using Distant Supervision. The existing approach requires huge computation resource for analysing large number of tweets. In this paper, we propose techniques to speed up the computation process for sentiment analysis. We use tweet subjectivity to select the right training samples. We also introduce the concept of EFWS (Effective Word Score) of a tweet that is derived from polarity scores of frequently used words, which is an additional heuristic that can be used to speed up the sentiment classification with standard machine learning algorithms. We performed our experiments using 1.6 million tweets. Experimental evaluations show that our proposed technique is more efficient and has higher accuracy compared to previously proposed methods. We achieve overall accuracies of around 80% (EFWS heuristic gives an accuracy around 85%) on a training dataset of 100K tweets, which is half the size of the dataset used for the baseline model. The accuracy of our proposed model is 2-3% higher than the baseline model, and the model effectively trains at twice the speed of the baseline model.