ENCLP 4 (ACL 2021)

NLP and IR have been closely linked to e-commerce applications since the early days of the fields. This close relationship between the two is evidenced by early publications as well as the growing number of recent publications at the intersection of the two areas.

Today, NLP and IR play a significant role in e-commerce tasks, including product search, recommender systems, product question answering, sentiment analysis, product description and review summarization, chatbots and shopping assistants, and customer review processing, among many other tasks being investigated by researchers in the field. These methods play a key part in today's online retail and shopping landscape, and continue to evolve and further enhance the customer experience.

The ECNLP workshop aims to provide a venue for the dissemination of NLP/IR research related to e-commerce and online shopping, bringing together researchers from both academia and industry.

The workshop will be held virtually on Friday August 6, 2021 (9:00 AM to 8:15 PM EDT).

You can join us via Zoom: https://zoom.us/j/91778858275?pwd=SURwVHV3V3NTcW1oVGQyQmY5UXpKQT09

• Meeting passcode: 881478

• Zoom instructions: https://support.zoom.us/hc/en-us/articles/201362193-joining-a-meeting

Proceedings: https://aclanthology.org/volumes/2021.ecnlp-1/

Why did Rachel break up with Ross? Deep Movie QA with Multimedia Knowledge Graph Construction

Speaker: Prof. Heng Ji (University of Illinois at Urbana-Champaign)

Just like book clubs, movie clubs offer an opportunity to gather with friends for fun discussion and deep understanding of a movie. When our friends are too busy we would like to have a movie club with our favorite virtual assistants. Existing virtual assistants (VAs) can answer superficial questions with limited knowledge from a movie’s meta-data information about cast, or its brief text summary about the content. However, they generally lack capabilities to respond to user inquiries in the movie domain with deep knowledge and engaging multimedia interactions, the latter being especially valuable for an enriched user experience that takes advantage of the touchscreen functionality which is available in some VA devices. In this talk, I will present a new system that can not only answer questions about content of movies and TV shows, but also retrieve relevant keyframes from the video to supplement the answer. In contrast to the existing work that mostly focuses on individual scene understanding, our system is based on a novel multi-hop, multimedia knowledge graph based question answering (QA) Framework for long-distance video understanding, in order to handle complex questions like “How did Andy and Red become friends in The Shawshank Redemption?”.

Using AI to Understand Search Intent

Speaker: Aritra Mandal (eBay Inc.)

In this talk, I will explore how AI can help understand query intent in the context of eCommerce search. We will focus on two specific aspects of this problem: query categorization and query similarity. Specifically, I will describe how to train a query categorization model from engagement data and how to recognize equivalent queries using embeddings trained from search behavior.

Understanding Query Intent in Shopping

Speaker: Corby Rosset (You.com)

Understanding the intent of users’ queries in online shopping is the most important first step to narrowing the search space to return relevant products. We will discuss how to leverage language models and large scale user clicks from search logs to weakly supervise a “query intent” encoder, which maps queries with shared clicks into similar embeddings end-to-end. Experimental results on an intrinsic evaluation task – query intent similarity modeling – demonstrate these methods' advantages over previous representation methods. Ablation studies reveal the crucial role of learning from implicit user feedback in representing user intent and the contributions of multi-task learning in representation generality. We also demonstrate the distributed representations alleviate the sparsity of tail search traffic and cuts down half of the unseen queries by using an efficient approximate nearest neighbor search to effectively identify previous queries with the same search intent. Finally, we demonstrate distances between GEN encodings reflect certain information seeking behaviors in search sessions.

Gazetteer Enhanced Named Entity Recognition for Code-Mixed Web Queries

Speaker: Besnik Fetahu (Amazon)

Named entity recognition (NER) for Web queries is very challenging. Queries often do not consist of well-formed sentences, and contain very little context, with queried entities being highly ambiguous. Code-mixed queries, which contain entities from a different language than the rest of the query, pose a particular challenge in domains like e-commerce (e.g. movie or product names). In this talk, we will discuss about NER approaches for code-mixed queries, where entities and non-entity query terms co-exist simultaneously in different languages. More specifically, we discuss about two aspects: (i) dataset generation for code-mixed NER, and (ii) an approach for handling code-mixed queries for NER. For (i), to address the lack of code-mixed NER data we will discuss on how we create EMBER, a large-scale dataset in six languages with four different scripts. Based on Bing query data, we include numerous language combinations that showcase real-world search scenarios. Secondly, we will discuss a novel gated architecture that enhances existing multi-lingual Transformers with a Mixture-of-Experts model to dynamically infuse multi-lingual gazetteers, allowing it to simultaneously differentiate and handle entities and non-entity query terms in multiple languages. Finally, we will discuss extensive experimental evaluation, where the utility of EMBER and the proposed approach is highlighted.

• Submission Deadline: May 7, 2021 (AoE) April 26, 2021

• Acceptance Notification: May 28, 2021

• Camera-ready versions: June 7, 2021

• Workshop: August 6, 2021

ECNLP 4 invites quality research contributions in different formats:

• • Original long research papers (8 pages plus references and appendix)

  • Original short research papers (4 pages plus references and appendix)

  • Position and opinion papers (4 pages)

ECNLP will use the ACL Submission Guidelines. Details and templates available here: https://2021.aclweb.org/calls/papers/

Submission website: https://www.softconf.com/acl2021/w08_ECNLP4/