31 January 2023 in ACM Transactions on Information Systems
Existing aspect-based/category sentiment analysis methods have shown great success in detecting sentiment polarity toward a given aspect in a sentence with supervised learning, where the training and inference stages share the same pre-defined set of aspects. However, in practice, the aspect categories are changing rather than keeping fixed over time. Dealing with unseen aspect categories is under-explored in existing methods. In this article, we formulate a new few-shot aspect category sentiment analysis (FSACSA) task, which aims to effectively predict the sentiment polarity of previously unseen aspect categories. To this end, we propose a novel Aspect-Focused Meta-Learning (AFML) framework that constructs aspect-aware and aspect-contrastive representations from external knowledge to match the target aspect with aspects in the training set. Concretely, we first construct two auxiliary contrastive sentences for a given sentence with the incorporation of external knowledge, enabling the learning of sentence representations with a better generalization. Then, we devise an aspect-focused induction network to leverage the contextual sentiment toward a given aspect to refine the label vectors. Furthermore, we employ the episode-based meta-learning algorithm to train the whole network, so as to learn to generalize to novel aspects. Extensive experiments on multiple real-life datasets show that our proposed AFML framework achieves the state-of-the-art results for the FSACSA task.
29 July 2020 in IEEE Access
Discovering frequent itemsets is a data analysis task used in numerous domains. It consists of finding sets of items (itemsets) that frequently appear in a set of database records (also called transactions). Though discovering frequent itemsets is useful, it can produce a large amount of spurious patterns. As a result, the user may spend a great amount of time to analyze the itemsets found by a frequent itemset mining algorithm to find truly interesting patterns. Hence, in recent years, a key research topic has emerged which is to discover statistically significant patterns in databases. The most popular model for identifying itemsets that are statistically significant is to discover non-redundant productive itemsets. The state-of-the-art algorithm to extract this set of patterns is OPUS-Miner. A key drawback of that algorithm is that it is designed to be applied to a static database. Moreover, a second drawback of OPUS-Miner is that it discovers all patterns in a database. In other words, the user cannot search for itemsets containing some specific items. This paper addresses these issues by defining the novel problem of discovering targeted non redundant productive itemsets in dynamic databases. An algorithm named IDPI+ (Interactive Discovery of Productive Itemsets) is presented, storing transactions in a tree structure, which can then be interactively queried to identify productive and non redundant itemsets containing specific items. A structure named Query-Tree is also introduced to process many queries at the same time. Moreover, to handle dynamic databases, efficient transaction insertion and deletion algorithms are provided to update the tree. It was observed in an experimental evaluation on benchmark datasets containing various types of data that IDPI+ can handle thousands of queries per second on a desktop computer. Moreover, it was found that IPDI+ is more than an order of magnitude faster than a baseline algorithm.