ICNLSP 2023: GAVI: A Category-Aware Generative Approach for Brand Value Identification, Kassem Sabeh

Title of the presentation: GAVI: A Category-Aware Generative Approach for Brand Value Identification.
By: Kassem Sabeh, Mouna Kacimi, Johann Gamper.
Free University of Bozen-Bolzano.
6th International Conference on Natural Language and Speech Processing.
icnlsp.org/2023welcome
Abstract:
Extracting product attribute value information is vital for many e-commerce applications. One
of the most crucial product attributes is the brand, as it significantly impacts customers’
purchasing decisions and behaviour. Consequently, it is critical for e-commerce platforms
to automatically and accurately identify brand values from product descriptions. Most existing methods focus on brand value extraction from text descriptions using sequence tagging
and question answering techniques. However, brand values are often not mentioned explicitly in the product descriptions. Also, these approaches are designed without paying attention to product categories, which are important for brand value identification. In this work, we
propose a novel category-aware generative approach for brand value identification (GAVI).
In particular, we formulate the brand value identification problem as a sequence-to-sequence
generation task. We use the T5 language model as the backbone of our approach. This allows
us to identify brand values that are not explicitly mentioned in the title in a generative manner.
We then propose to highlight the product categories inside our model input, making the approach category-aware. We conduct extensive experiments on a public dataset for brand value
identification. The experimental results demonstrate that our generation-based approach outperforms existing extraction-based methods.
Our code is released along with the fine-tuned models presented in the paper , which are also
available as a demo.
.