Product Classification Using Microdata Annotations

An Exploratory Study on Utilising the Web of Linked Data for Product Data Mining

“…In addition, Ristoski et al (2018) also used product images. The work by Meusel et al (2015) and Zhang and Paramita (2019) used product categories allocated by the vendors and embedded as semantic markup data within the web pages. To differentiate these from the classification targets in such tasks, we refer to these as 'site-specific product labels' or 'categories'.…”

“…Certain techniques will need to be applied in order to compose embeddings for long text passages based on single words. For example, Kozareva (2015) averaged the embedding vectors of composing words from product titles, Lee and Yoon (2018) summed them, while work in Kim (2014) and Zhang and Paramita (2019) joined word embedding vectors to create a 2D tensor to represent the text. In the more recent work that uses pre-trained LMs such as BERT (e.g.…”

“…The large majority of work has used supervised machine learning methods. These include those that use traditional machine learning algorithms (Meusel et al (2015); Cevahir and Murakami (2016); Chavaltada et al (2017); Gupta et al (2016); Lee and Yoon (2018); Ristoski et al (2018)), and those that apply DNN-based algorithms (Xia et al (2017); Ha et al (2016); Lin et al (2018); Zhang and Paramita (2019)). All DNN-based methods have used some adaptations of CNN or RNN.…”

“…Second, despite the continued interest in the research of product classification and linking tasks, the use of language resources to support such tasks has been highly inconsistent, ranging from no-use at all (e.g., Cevahir and Murakami (2016) and many described in Lin et al (2018)), to diverse ways of using a diverse set of word embedding models (e.g., Kozareva (2015); Zhang and Paramita (2019); Borst et al (2020)). Adding to this complexity is the use of different datasets, and diverse use of machine learning models ranging from traditional algorithms (e.g., SVM, logistic regression), to deep neural networks, to pre-trained language models.…”

“…For the DNN-based methods, we use the 'GN-DeepCN' structure proposed in Zhang and Paramita (2019), where multiple channels take input of a specific type of product metadata and learn separate feature representation for them, and finally concatenate these different representations into a single feature vector for classification. Specifically, we take their 'bi-LSTM' (GND.biLSTM) and 'HAN' (GND.HAN) versions where each channel is implemented as a bi-LSTM and HAN structure respectively.…”