Yahoo! Trending Now lists the most trending ten user queries from Yahoo! Search. To discover top trending queries, query clustering is a critical intermediate phase that aggregates similar queries into clusters, each representing an event or a topic. Established on a heuristic clustering framework, the existing approach can generate suboptimal results but lacks the ability to: 1) fully exploit semantic information in news articles associated with user queries, and 2) account for changes in queries and news articles over consecutive timestamps. In this paper, we first introduce a two-stage query clustering framework that leverages both match-based grouping and distance-based clustering. This novel and effective solution significantly surpasses the existing production method. Furthermore, to address the challenges posed by high time complexity and potential cluster fluctuations on account of temporal factors, we optimize the newly proposed framework by 1) utilizing a caching mechanism to store historical query features to enhance computational efficiency, and 2) applying voting and rolling average strategies at the time window level to both stages, respectively, resulting in smoother feature representations and more robust clustering out-comes. Through offline evaluation, our integrated method speeds up the baseline by 20 times and reduces cluster fluctuations by 15 times. These improvements considerably enhance the efficiency and stableness of query clustering for Yahoo! Trending Now.

Unsupervised aspect detection (UAD) aims at automatically extracting interpretable aspects and identifying aspect-specific segments (such as sentences) from online reviews. However, recent deep learning based topic models, specifically aspect-based autoencoder, suffer from several problems such as extracting noisy aspects and poorly mapping aspects discovered by models to the aspects of interest. To tackle these challenges, in this paper, we first propose a self-supervised contrastive learning framework and an attention-based model equipped with a novel smooth self-attention (SSA) module for the UAD task in order to learn better representations for aspects and review segments. Secondly, we introduce a high-resolution selective mapping (HRSMap) method to efficiently assign aspects discovered by the model to the aspects of interest. We also propose using a knowledge distillation technique to further improve the aspect detection performance. Our methods outperform several recent unsupervised and weakly supervised approaches on publicly available benchmark user review datasets. Aspect interpretation results show that extracted aspects are meaningful, have a good coverage, and can be easily mapped to aspects of interest. Ablation studies and attention weight visualization also demonstrate effectiveness of SSA and the knowledge distillation method.

Programmers produce code clones when developing software. By copying and pasting code with or without modification, developers reuse existing code to improve programming productivity. However, code clones present challenges to software maintenance: they may require consistent application of the same or similar bug fixes or program changes to multiple code locations. To simplify the maintenance process, various tools have been proposed to automatically detect clones [1], [2], [3], [4], [5], [6]. Some tools tokenize source code, and then compare the sequence or frequency of tokens to reveal clones [1], [3], [4], [5]. Some other tools detect clones using tree-matching algorithms to compare the Abstract Syntax Trees (ASTs) of source code [2], [6]. In this paper, we present CCLEARNER, the first solely token-based clone detection approach leveraging deep learning. CCLEARNER extracts tokens from known method-level code clones and nonclones to train a classifier, and then uses the classifier to detect clones in a given codebase. To evaluate CCLEARNER, we reused BigCloneBench [7], an existing large benchmark of real clones. We used part of the benchmark for training and the other part for testing, and observed that CCLEARNER effectively detected clones. With the same data set, we conducted the first systematic comparison experiment between CCLEARNER and three popular clone detection tools. Compared with the approaches not using deep learning, CCLEARNER achieved competitive clone detection effectiveness with low time cost.