Papers by Besnik Fetahu
Reinforced Question Rewriting for Conversational Question Answering (2022.emnlp-industry)
Copied to clipboard
| Challenge: | Existing approaches to CQA involve training new models from scratch . existing approaches are expensive and often not feasible . |
| Approach: | They propose to use QA feedback to supervise the rewriting model with reinforcement learning. |
| Outcome: | The proposed model can improve QA performance over baselines for extractive and retrieval QA. |
Dynamic Gazetteer Integration in Multilingual Models for Cross-Lingual and Cross-Domain Named Entity Recognition (2022.naacl-main)
Copied to clipboard
| Challenge: | Named entity recognition (NER) models trained on CoNLL do not transfer well to other domains, even within the same language. |
| Approach: | They propose a token-level gating layer to augment pre-trained multilingual transformers with gazetteers containing named entities (NE) from a target language or domain. |
| Outcome: | The proposed model improves on cross-lingual transfer with an F1 score of 92.92 for English and an average of 89.43 across all languages in CoNLL. |
Distilling Multilingual Transformers into CNNs for Scalable Intent Classification (2022.emnlp-industry)
Copied to clipboard
| Challenge: | Existing multilingual models for voice assistants are limited by their prohibitive inference time and limited performance. |
| Approach: | They propose to distill and deploy multilingual Transformer models for voice assistants using a teacher-student framework that uses teacher-trained models to supervise student model training. |
| Outcome: | The proposed model outperforms a teacher model trained on unlabelled data and achieves equivalent performance. |
InstructPTS: Instruction-Tuning LLMs for Product Title Summarization (2023.emnlp-industry)
Copied to clipboard
| Challenge: | E-commerce product catalogs contain billions of items with lengthy titles . this leads to a gap between how customers refer to these unnatural titles - and how they are used . |
| Approach: | They propose a novel approach to product title summarization that uses a fine-tuned instruction strategy to train a highly accurate model. |
| Outcome: | The proposed approach can generate more accurate product title summaries with an improvement of over 14 and 8 BLEU and ROUGE points. |
Generative Explore-Exploit: Training-free Optimization of Generative Recommender Systems using LLM Optimizers (2024.acl-long)
Copied to clipboard
Lütfi Kerem Senel, Besnik Fetahu, Davis Yoshida, Zhiyu Chen, Giuseppe Castellucci, Nikhita Vedula, Jason Ingyu Choi, Shervin Malmasi
| Challenge: | Large Language Models (LLMs) have given rise to generative recommenders . however, improving the generated content through user feedback is prohibitively expensive . |
| Approach: | They propose a generative explore-exploit method that exploits items with high engagement and actively explores hidden population preferences to improve recommendation quality. |
| Outcome: | The proposed approach exploits items with high engagement and actively explores hidden population preferences to improve recommendation quality. |
MultiCoNER: A Large-scale Multilingual Dataset for Complex Named Entity Recognition (2022.coling-1)
Copied to clipboard
| Challenge: | Named Entity Recognition (NER) is a core task in Natural Language Processing. |
| Approach: | They present a large multilingual dataset for Named Entity Recognition that covers 3 domains across 11 languages and multilingual and code-mixing subsets. |
| Outcome: | The proposed dataset is large and multilingual, covering 11 languages and subsets. |
SDR: Efficient Neural Re-ranking using Succinct Document Representation (2022.acl-long)
Copied to clipboard
| Challenge: | BERT based ranking models have been successful on various information retrieval tasks, but they are prone to storage and network fetching latency. |
| Approach: | They propose a late-interaction architecture that allows pre-computation of intermediate document representations, thus reducing latency. |
| Outcome: | The proposed model achieves 4x–11.6x higher compression rates on the MSMARCO passage re-reranking task compared to existing methods. |
Answering Unanswered Questions through Semantic Reformulations in Spoken QA (2023.acl-industry)
Copied to clipboard
| Challenge: | Question Answering (QA) is a longstanding NLP task, and voice assistants like Alexa have made Spoken QA ubiquitous. |
| Approach: | They propose a model that uses linguistically-grounded operations to rewrite questions to facilitate answering. |
| Outcome: | The proposed model improves answer rates on 1M unanswered questions from a leading voice assistant. |
Neural OCR Post-Hoc Correction of Historical Corpora (2021.tacl-1)
Copied to clipboard
| Challenge: | Optical character recognition (OCR) is crucial for a deeper access to historical collections. |
| Approach: | They propose a neural approach based on a combination of recurrent (RNN) and deep convolutional network (ConvNet) to correct OCR transcription errors. |
| Outcome: | The proposed model reduces the word error rate of 32.3% by more than 89% on a historical book corpus in German language. |
Wizard of Shopping: Target-Oriented E-commerce Dialogue Generation with Decision Tree Branching (2025.acl-long)
Copied to clipboard
Xiangci Li, Zhiyu Chen, Jason Ingyu Choi, Nikhita Vedula, Besnik Fetahu, Oleg Rokhlenko, Shervin Malmasi
| Challenge: | Prior human-annotated CPS datasets are small in size and lack integration with real-world product search systems. |
| Approach: | They propose a method to generate target-oriented shopping conversations without human annotations by using large language models. |
| Outcome: | The proposed method achieves highly natural and coherent conversations from three shopping domains and significantly improves on human evaluations and downstream tasks. |
MultiCoNER v2: a Large Multilingual dataset for Fine-grained and Noisy Named Entity Recognition (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Named Entity Recognition (NER) is a core task in Natural Language Processing. |
| Approach: | They present a dataset for fine-grained Named Entity Recognition covering 33 entity classes across 12 languages in monolingual and multilingual settings. |
| Outcome: | The proposed dataset covers 33 entity classes across 12 languages in monolingual and multilingual settings. |
CycleKQR: Unsupervised Bidirectional Keyword-Question Rewriting (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to query paraphrases are based on encoderdecoder architectures, but they do not support the two important functionalities beyond questions. |
| Approach: | They propose a keyword-question rewriting task to improve query understanding capabilities of NLU systems for all surface forms. |
| Outcome: | Empirically, we show that CycleKQR significantly improves QA performance by rewriting queries into the appropriate form while retaining the original semantic meaning of input queries. |
Follow-on Question Suggestion via Voice Hints for Voice Assistants (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Query suggestion is a standard feature of screen-based search experiences, but it is not trivial to implement in voice-based settings. |
| Approach: | They propose a task of suggesting questions with compact voice hints to allow users to ask follow-up questions. |
| Outcome: | The proposed approach is based on a dataset of 6681 input questions and human written hints and is highly linguistically motivated. |
Generate-then-Retrieve: Intent-Aware FAQ Retrieval in Product Search (2023.acl-industry)
Copied to clipboard
| Challenge: | Frequently Asked Question (FAQ) retrieval aims at retrieving question-answer pairs for a given user query. |
| Approach: | They propose to use an intent classifier to predict whether a query is looking for an FAQ and a reformulation model to rewrite the query into a natural question to improve retrieval performance. |
| Outcome: | The proposed method improves 12% on Hit@1 on retrieving ground-truth FAQs while reducing latency by 95% compared to baseline systems. |
Identifying High Consideration E-Commerce Search Queries (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Identifying high consideration queries is essential for e-commerce sites to better serve user needs . ecommerce sites can create or serve customized content for specific queries . |
| Approach: | They propose an engagement-based Query Ranking approach to identify potential engagement levels with query-related shopping knowledge content during product search. |
| Outcome: | The proposed method outperforms human-selected queries in terms of customer impact . human evaluation shows a precision of 96% for HC queries identified by the model . |