Abstract
As the amount of generated information grows, reading and summarizing texts of large collections turns into a challenging task. Many documents do not come with descriptive terms, thus requiring humans to generate keywords on-the-fly. The need to automate this kind of task demands the development of keyword extraction systems with the ability to automatically identify keywords within the text. One approach is to resort to machine-learning algorithms. These, however, depend on large annotated text corpora, which are not always available. An alternative solution is to consider an unsupervised approach. In this article, we describe YAKE!, a light-weight unsupervised automatic keyword extraction method which rests on statistical text features extracted from single documents to select the most relevant keywords of a text. Our system does not need to be trained on a particular set of documents, nor does it depend on dictionaries, external corpora, text size, language, or domain. To demonstrate the merits and significance of YAKE!, we compare it against ten state-of-the-art unsupervised approaches and one supervised method. Experimental results carried out on top of twenty datasets show that YAKE! significantly outperforms other unsupervised methods on texts of different sizes, languages, and domains. © 2019 Elsevier Inc.

Abstract

Abstract

Abstract
Users tend to search over the Internet to get the most updated news when an event occurs. Search engines should then be capable of effectively retrieving relevant documents for event-related queries. As the previous studies have shown, different retrieval models are needed for different types of events. Therefore, the first step for improving effectiveness is identifying the event-related queries and determining their types. In this paper, we propose a novel model based on deep neural networks to classify event-related queries into four categories: periodic, aperiodic, one-time-only, and non-event. The proposed model combines recurrent neural networks (by feeding two LSTM layers with query frequencies) and visual recognition models (by transforming time-series data from a 1D signal to a 2D image - later passed to a CNN model) for effective query type estimation. Worth noting is that our method uses only the time-series data of query frequencies, without the need to resort to any external sources such as contextual data, which makes it language and domain-independent with regards to the query issued. For evaluation, we build upon the previous datasets on event-related queries to create a new dataset that fits the purpose of our experiments. The obtained results show that our proposed model can achieve an F1-score of 0.87. © 2020 ACM.

Abstract