Abstract:We propose a novel method for counting sentiment orientation that outperforms supervised learning approaches in time and memory complexity and is not statistically significantly different from them in accuracy. Our method consists of a novel approach to generating unigram, bigram and trigram lexicons. The proposed method, called frequentiment, is based on calculating the frequency of features (words) in the document and averaging their impact on the sentiment score as opposed to documents that do not contain these features. Afterwards, we use ensemble classification to improve the overall accuracy of the method. What is important is that the frequentiment-based lexicons with sentiment threshold selection outperform other popular lexicons and some supervised learners, while being 3-5 times faster than the supervised approach. We compare 37 methods (lexicons, ensembles with lexicon's predictions as input and supervised learners) applied to 10 Amazon review data sets and provide the first statistical comparison of the sentiment annotation methods that include ensemble approaches. It is one of the most comprehensive comparisons of domain sentiment analysis in the literature.

No abstract

Natural language processing of conversational speech requires the availability of high-quality transcripts. In this paper, we express our skepticism towards the recent reports of very low Word Error Rates (WERs) achieved by modern Automatic Speech Recognition (ASR) systems on benchmark datasets. We outline several problems with popular benchmarks and compare three state-of-the-art commercial ASR systems on an internal dataset of real-life spontaneous human conversations and HUB'05 public benchmark. We show that WERs are significantly higher than the best reported results. We formulate a set of guidelines which may aid in the creation of real-life, multi-domain datasets with high quality annotations for training and testing of robust ASR systems.

We proposed a new accurate aspect extraction method that makes use of both word and characterbased embeddings. We have conducted experiments of various models of aspect extraction using LSTM and BiLSTM including CRF enhancement on five different pre-trained word embeddings extended with character embeddings. The results revealed that BiLSTM outperforms regular LSTM, but also word embedding coverage in train and test sets profoundly impacted aspect detection performance. Moreover, the additional CRF layer consistently improves the results across different models and text embeddings. Summing up, we obtained state-of-the-art F-score results for SemEval Restaurants (85%) and Laptops (80%).