Data Fusion for Better Fake Reviews Detection
20034Authors
nan
Abstract
AbstractOnline reviews have become critical in informing purchasing decisions, making the detection of fake reviews a crucial challenge to tackle. Many different Machine Learning based solutions have been proposed, using various data representations such as n-grams or document embeddings. In this paper, we first explore the effectiveness of different data representations, including emotion, document embedding, n-grams, and noun phrases in embedding for mat, for fake reviews detection. We evaluate these representations with various state-of-the-art deep learning models, such as BILSTM, LSTM, GRU, CNN, and MLP. Following this, we propose to incorporate different data repre- sentations and classification models using early and late data fusion techniques in order to im- prove the prediction performance. The experiments are conducted on four datasets: Hotel, Restaurant, Amazon, and Yelp. The results demonstrate that combination of different data representations significantly outperform any of the single data representations
Keywords
Full Content
Proceedings of Recent Advances in Natural Language Processing, pages 60–68 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_007 60 Bhojpuri WordNet: Problems in Translating Hindi Synsets into Bhojpuri Imran Ali Banaras Hindu University Varanasi, India aravimran@bhu.ac.in Praveen Gatla Banaras Hindu University Varanasi, India praveengatla@bhu.ac.in Abstract Today, artificial intelligence systems are incred- ibly intelligent, however, they lack the human- like capacity for understanding. In this context, sense-based lexical resources become a require- ment to develop artificial intelligent machines. Lexical resources like Wordnets have received scholarly attention because they are considered crucial sense-based resources in the field of nat- ural language understanding. They can help the machines in knowing the intended meaning of the communicated texts, as they are focused on the concept rather than the words. Word- nets are available only for 18 Indian languages. Keeping this in mind, we have initiated the development of a comprehensive wordnet for Bhojpuri. The present paper describes the cre- ation of the synsets of Bhojpuri and discusses the problems that we faced while translating Hindi synsets into Bhojpuri. Some of the chal- lenges are lexical anomalies, lexical mismatch words, synthesized forms, lack of technical words, etc. Nearly 4000 Hindi synsets were mapped for their equivalent synsets in Bhojpuri by following the expansion approach. We have also worked on the language-specific synsets, which are unique to Bhojpuri. This resource is useful in machine translation, sentiment analy- sis, word sense disambiguation, cross-lingual references among Indian languages, and Bho- jpuri language teaching and learning. 1 Introduction Today’s era is one of science and technology. People have been communicating using the Internet and social media and enjoying different forms of media and entertainment. For this, they require accessible resources in their own languages; however, we Indians are forced to depend on the tools that are available, either in English or only in a few major Indian languages. The creation of linguistic resources in a language, particularly in a low-resourced language, is a very challenging task. To understand the intended meaning of a communicated text, one needs knowledge of the world along with competency in the language, which cannot be captured with any traditional resources as meaning resides not in the words but in the minds of the people using them (Nida, 1979). We need a very comprehensive and intelligent tool to understand a text like a human. In recent years, wordnets have been considered a very crucial tool in the field of natural language processing. WordNet is an online lexical resource and a semantic network (Bhattacharyya, 2010). It is constituted of synsets. Each synset expresses a distinct concept. So synsets are the basic building blocks of WordNet (Bhattacharyya et al., 2006). WordNet’s design is inspired by the current psycholinguistic theories of human lexical memory (Miller, 1998). WordNet stores lexical items in ontological that are used to represent IS-A-KIND- OF, IS-A-PART-OF and other relations such as the hypernymy-hyponymy and holonymy-meronymy. Wordnets have been developed for more than 200 languages (Rebele et al., 2016) because wordnets are considered to be the most important …
Content truncated to 500 words
Labeled Text
AbstractOnline reviews have become critical in informing purchasing decisions, making the detection of fake reviews a crucial challenge to tackle. Many different Machine Learning based solutions have been proposed, using various data representations such as n-grams or document embeddings. In this paper, we first explore the effectiveness of different data representations, including emotion, document embedding, n-grams, and noun phrases in embedding for mat, for fake reviews detection. We evaluate these representations with various state-of-the-art deep learning models, such as BILSTM, LSTM, GRU, CNN, and MLP. Following this, we propose to incorporate different data repre- sentations and classification models using early and late data fusion techniques in order to im- prove the prediction performance. The experiments are conducted on four datasets: Hotel, Restaurant, Amazon, and Yelp. The results demonstrate that combination of different data representations significantly outperform any of the single data representations
Publication Details
- Paper ID:
- 20034
- Year:
- 2023
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan