In recent years, end-to-end automatic speech recognition (ASR) systems have proven themselves remarkably accurate and performant, but these systems still have a significant error rate for entity names which appear infrequently in their training data. In parallel to the rise of end-to-end ASR systems, large language models (LLMs) have proven to be a versatile tool for various natural language processing (NLP) tasks. In NLP tasks where a database of relevant knowledge is available, retrieval augmented generation (RAG) has achieved impressive results when used with LLMs. In this work, we propose a RAG-like technique for correcting speech recognition entity name errors. Our approach uses a vector database to index a set of relevant entities. At runtime, database queries are generated from possibly errorful textual ASR hypotheses, and the entities retrieved using these queries are fed, along with the ASR hypotheses, to an LLM which has been adapted to correct ASR errors. Overall, our best system achieves 33%-39% relative word error rate reductions on synthetic test sets focused on voice assistant queries of rare music entities without regressing on the STOP test set, a publicly available voice assistant test set covering many domains.

Directed graphs have been widely used in Community Question Answering services (CQAs) to model asymmetric relationships among different types of nodes in CQA graphs, e.g., question, answer, user. Asymmetric transitivity is an essential property of directed graphs, since it can play an important role in downstream graph inference and analysis. Question difficulty and user expertise follow the characteristic of asymmetric transitivity. Maintaining such properties, while reducing the graph to a lower dimensional vector embedding space, has been the focus of much recent research. In this paper, we tackle the challenge of directed graph embedding with asymmetric transitivity preservation and then leverage the proposed embedding method to solve a fundamental task in CQAs: how to appropriately route and assign newly posted questions to users with the suitable expertise and interest in CQAs. The technique incorporates graph hierarchy and reachability information naturally by relying on a nonlinear transformation that operates on the core reachability and implicit hierarchy within such graphs. Subsequently, the methodology levers a factorization-based approach to generate two embedding vectors for each node within the graph, to capture the asymmetric transitivity. Extensive experiments show that our framework consistently and significantly outperforms the state-of-the-art baselines on three diverse realworld tasks: link prediction, and question difficulty estimation and expert finding in online forums like Stack Exchange. Particularly, our framework can support inductive embedding learning for newly posted questions (unseen nodes during training), and therefore can properly route and assign these kinds of questions to experts in CQAs.

Sparse Matrix-Vector (SpMV) and Sparse Matrix-Multivector (SpMM) products are key kernels for computational science and data science. While GPUs offer significantly higher peak performance and memory bandwidth than multicore CPUs, achieving high performance on sparse computations on GPUs is very challenging. A tremendous amount of recent research has focused on various GPU implementations of the SpMV kernel. But the multi-vector SpMM kernel has received much less attention. In this paper, we present an in-depth analysis to contrast SpMV and SpMM, and develop a new sparse-matrix representation and computation approach suited to achieving high data-movement efficiency and effective GPU parallelization of SpMM. Experimental evaluation using the entire SuiteSparse matrix suite demonstrates significant performance improvement over existing SpMM implementations from vendor libraries.

We propose a novel, scalable, and principled graph sketching technique based on minwise hashing of local neighborhood. For an n-node graph with e-edges (e >> n), we incrementally maintain in real-time a minwise neighbor sampled subgraph using k hash functions in O(n x k) memory, limit being user-configurable by the parameter k. Symmetrization and similarity based techniques can recover from these data structures a significant portion of the original graph. We present theoretical analysis of the minwise sampling strategy and also derive unbiased estimators for important graph properties such as triangle count and neighborhood overlap.

Social media applications have assumed an important role in decision-making process of users, affecting their choices about products and services. In this context, understanding and modeling opinions, as well as opinion-leaders, have implications for several tasks, such as recommendation, advertising, brand evaluation etc. Despite the intrinsic relation between opinions and opinion-leaders, most recent works focus exclusively on either understanding the opinions, by Sentiment Analysis (SA) proposals, or identifying opinion-leaders using Influential Users Detection (IUD). This paper presents a preliminary evaluation about a combined analysis of SA and IUD. In this sense, we propose a methodology to quantify factors in real domains that may affect such analysis, as well as the potential benefits of combining SA Methods with IUD ones. Empirical assessments on a sample of tweets about the Brazilian president reveal that the collective opinion and the set of top opinion-leaders over time are inter-related. Further, we were able to identify distinct characteristics of opinion propagation, and that the collective opinion may be accurately estimated by using a few top-k opinion-leaders. These results point out the combined analysis of SA and IUD as a promising research direction to be further exploited.