Eugene Agichtein

Anna Gottardi, Osman Ipek, Giuseppe Castellucci et al. · amazon-science

Since its inception in 2016, the Alexa Prize program has enabled hundreds of university students to explore and compete to develop conversational agents through the SocialBot Grand Challenge. The goal of the challenge is to build agents capable of conversing coherently and engagingly with humans on popular topics for 20 minutes, while achieving an average rating of at least 4.0/5.0. However, as conversational agents attempt to assist users with increasingly complex tasks, new conversational AI techniques and evaluation platforms are needed. The Alexa Prize TaskBot challenge, established in 2021, builds on the success of the SocialBot challenge by introducing the requirements of interactively assisting humans with real-world Cooking and Do-It-Yourself tasks, while making use of both voice and visual modalities. This challenge requires the TaskBots to identify and understand the user's need, identify and integrate task and domain knowledge into the interaction, and develop new ways of engaging the user without distracting them from the task at hand, among other challenges. This paper provides an overview of the TaskBot challenge, describes the infrastructure support provided to the teams with the CoBot Toolkit, and summarizes the approaches the participating teams took to overcome the research challenges. Finally, it analyzes the performance of the competing TaskBots during the first year of the competition.

Chen Lin, Safoora Yousefi, Elvis Kahoro et al. · amazon-science

Real-time air pollution monitoring is a valuable tool for public health and environmental surveillance. In recent years, there has been a dramatic increase in air pollution forecasting and monitoring research using artificial neural networks (ANNs). Most of the prior work relied on modeling pollutant concentrations collected from ground-based monitors and meteorological data for long-term forecasting of outdoor ozone, oxides of nitrogen, and PM2.5. Given that traditional, highly sophisticated air quality monitors are expensive and are not universally available, these models cannot adequately serve those not living near pollutant monitoring sites. Furthermore, because prior models were built on physical measurement data collected from sensors, they may not be suitable for predicting public health effects experienced from pollution exposure. This study aims to develop and validate models to nowcast the observed pollution levels using Web search data, which is publicly available in near real-time from major search engines. We developed novel machine learning-based models using both traditional supervised classification methods and state-of-the-art deep learning methods to detect elevated air pollution levels at the US city level, by using generally available meteorological data and aggregate Web-based search volume data derived from Google Trends. We validated the performance of these methods by predicting three critical air pollutants (ozone (O3), nitrogen dioxide (NO2), and fine particulate matter (PM2.5)), across ten major U.S. metropolitan statistical areas (MSAs) in 2017 and 2018.

Simone Filice, Jason Ingyu Choi, Giuseppe Castellucci et al. · amazon-science

Many Natural Language Generation (NLG) tasks aim to generate a single output text given an input prompt. Other settings require the generation of multiple texts, e.g., for Synthetic Traffic Generation (STG). This generation task is crucial for training and evaluating QA systems as well as conversational agents, where the goal is to generate multiple questions or utterances resembling the linguistic variability of real users. In this paper, we show that common NLG metrics, like BLEU, are not suitable for evaluating STG. We propose and evaluate several metrics designed to compare the generated traffic to the distribution of real user texts. We validate our metrics with an automatic procedure to verify whether they capture different types of quality issues of generated data; we also run human annotations to verify the correlation with human judgements. Experiments on three tasks, i.e., Shopping Utterance Generation, Product Question Generation and Query Auto Completion, demonstrate that our metrics are effective for evaluating STG tasks, and improve the agreement with human judgement up to 20% with respect to common NLG metrics. We believe these findings can pave the way towards better solutions for estimating the representativeness of synthetic text data.

Zihao Wang, Ali Ahmadvand, Jason Choi et al. · amazon-science

Open-domain conversational search (ODCS) aims to provide valuable, up-to-date information, while maintaining natural conversations to help users refine and ultimately answer information needs. However, creating an effective and robust ODCS agent is challenging. In this paper, we present a fully functional ODCS system, Ericson, which includes state-of-the-art question answering and information retrieval components, as well as intent inference and dialogue management models for proactive question refinement and recommendations. Our system was stress-tested in the Amazon Alexa Prize, by engaging in live conversations with thousands of Alexa users, thus providing empirical basis for the analysis of the ODCS system in real settings. Our interaction data analysis revealed that accurate intent classification, encouraging user engagement, and careful proactive recommendations contribute most to the users satisfaction. Our study further identifies limitations of the existing search techniques, and can serve as a building block for the next generation of ODCS agents.

Zihao Wang, Eugene Agichtein, Jinho Choi · amazon-science

Response ranking in dialogues plays a crucial role in retrieval-based conversational systems. In a multi-turn dialogue, to capture the gist of a conversation, contextual information serves as essential knowledge to achieve this goal. In this paper, we present a flexible neural framework that can integrate contextual information from multiple channels. Specifically for the current task, our approach is to provide two information channels in parallel, Fusing Conversation history and domain knowledge extracted from Candidate provenance (FCC), where candidate responses are curated, as contextual information to improve the performance of multi-turn dialogue response ranking. The proposed approach can be generalized as a module to incorporate miscellaneous contextual features for other context-oriented tasks. We evaluate our model on the MSDialog dataset widely used for evaluating conversational response ranking tasks. Our experimental results show that our framework significantly outperforms the previous state-of-the-art models, improving Recall@1 by 7% and MAP by 4%. Furthermore, we conduct ablation studies to evaluate the contributions of each information channel, and of the framework components, to the overall ranking performance, providing additional insights and directions for further improvements.

Kaustubh D. Dhole, Ramraj Chandradevan, Eugene Agichtein

While search is the predominant method of accessing information, formulating effective queries remains a challenging task, especially for situations where the users are not familiar with a domain, or searching for documents in other languages, or looking for complex information such as events, which are not easily expressible as queries. Providing example documents or passages of interest, might be easier for a user, however, such query-by-example scenarios are prone to concept drift, and are highly sensitive to the query generation method. This demo illustrates complementary approaches of using LLMs interactively, assisting and enabling the user to provide edits and feedback at all stages of the query formulation process. The proposed Query Generation Assistant is a novel search interface which supports automatic and interactive query generation over a mono-linguial or multi-lingual document collection. Specifically, the proposed assistive interface enables the users to refine the queries generated by different LLMs, to provide feedback on the retrieved documents or passages, and is able to incorporate the users' feedback as prompts to generate more effective queries. The proposed interface is a valuable experimental tool for exploring fine-tuning and prompting of LLMs for query generation to qualitatively evaluate the effectiveness of retrieval and ranking models, and for conducting Human-in-the-Loop (HITL) experiments for complex search tasks where users struggle to formulate queries without such assistance.

Kaustubh D. Dhole, Eugene Agichtein

Large language models (LLMs) are increasingly evaluated and sometimes trained using automated graders such as LLM-as-judges that output scalar scores or preferences. While convenient, these approaches are often opaque: a single score rarely explains why an answer is good or bad, which requirements were missed, or how a system should be improved. This lack of interpretability limits their usefulness for model development, dataset curation, and high-stakes deployment. Query-specific rubric-based evaluation offers a more transparent alternative by decomposing quality into explicit, checkable criteria. However, manually designing high-quality, query-specific rubrics is labor-intensive and cognitively demanding and not feasible for deployment. While previous approaches have focused on generating intermediate rubrics for automated downstream evaluation, it is unclear if these rubrics are both interpretable and effective for human users. In this work, we investigate whether LLMs can generate useful, instance-specific rubrics as compared to human-authored rubrics, while also improving effectiveness for identifying good responses. Through our systematic study on two rubric benchmarks, and on multiple few-shot and post-training strategies, we find that off-the-shelf LLMs produce rubrics that are poorly aligned with human-authored ones. We introduce a simple strategy, RubricRAG, which retrieves domain knowledge via rubrics at inference time from related queries. We demonstrate that RubricRAG can generate more interpretable rubrics both for similarity to human-authored rubrics, and for improved downstream evaluation effectiveness. Our results highlight both the challenges and a promising approach of scalable, interpretable evaluation through automated rubric generation.

Harshita Jagdish Sahijwani, Madhav Sigdel, Song Aslan et al.

Classifying the intent behind healthcare search queries is crucial for improving the delivery of online healthcare information. The intricate nature of medical search queries, coupled with the limited availability of high-quality labeled data, presents substantial challenges for developing efficient classification models. Previous studies have exploited user interaction data, such as user clicks from search logs and employed pairwise loss functions to model co-click behavior for query representation learning. However, many health queries could have multiple intents, resulting in ambiguous or divergent click behavior. Furthermore, learning the single most popular intent of queries as inferred from global statistics based on the aggregate behavior of different users could potentially lead to disparity and performance drop when classifying the query intent within specific search sessions. To address these limitations, our work improves the query representation learning by aggregating similar queries via clustering, and introducing a novel loss function designed to capture the multifaceted nature of health search queries, resulting in a more scalable and accurate learning procedure. Furthermore, we quantify the ambiguity of health queries and the misalignment between global search intents and those discerned from individual sessions, by introducing the concordance rate (CR) score, and demonstrate a simple and effective method for incorporating our learned query representation into contextual, session-based search intent classification. Our extensive experimental results and analysis on two real-world search log datasets, i.e., a Health Search (HS) dataset and the publicly available TripClick dataset, demonstrate that our approach not only improves the intrinsic clustering metrics for query representation learning but also enhances accuracy for subsequent search intent classification tasks.

Jianghong Zhou, Joyce C. Ho, Chen Lin et al.

Interactive search can provide a better experience by incorporating interaction feedback from the users. This can significantly improve search accuracy as it helps avoid irrelevant information and captures the users' search intents. Existing state-of-the-art (SOTA) systems use reinforcement learning (RL) models to incorporate the interactions but focus on item-level feedback, ignoring the fine-grained information found in sentence-level feedback. Yet such feedback requires extensive RL action space exploration and large amounts of annotated data. This work addresses these challenges by proposing a new deep Q-learning (DQ) approach, DQrank. DQrank adapts BERT-based models, the SOTA in natural language processing, to select crucial sentences based on users' engagement and rank the items to obtain more satisfactory responses. We also propose two mechanisms to better explore optimal actions. DQrank further utilizes the experience replay mechanism in DQ to store the feedback sentences to obtain a better initial ranking performance. We validate the effectiveness of DQrank on three search datasets. The results show that DQRank performs at least 12% better than the previous SOTA RL approaches. We also conduct detailed ablation studies. The ablation results demonstrate that each model component can efficiently extract and accumulate long-term engagement effects from the users' sentence-level feedback. This structure offers new technologies with promised performance to construct a search system with sentence-level interaction.

Kaustubh Dhole, Eugene Agichtein · amazon-science

Query Reformulation(QR) is a set of techniques used to transform a user's original search query to a text that better aligns with the user's intent and improves their search experience. Recently, zero-shot QR has been shown to be a promising approach due to its ability to exploit knowledge inherent in large language models. By taking inspiration from the success of ensemble prompting strategies which have benefited many tasks, we investigate if they can help improve query reformulation. In this context, we propose an ensemble based prompting technique, GenQREnsemble which leverages paraphrases of a zero-shot instruction to generate multiple sets of keywords ultimately improving retrieval performance. We further introduce its post-retrieval variant, GenQREnsembleRF to incorporate pseudo relevant feedback. On evaluations over four IR benchmarks, we find that GenQREnsemble generates better reformulations with relative nDCG@10 improvements up to 18% and MAP improvements upto 24% over the previous zero-shot state-of-art. On the MSMarco Passage Ranking task, GenQREnsembleRF shows relative gains of 5% MRR using pseudo-relevance feedback, and 9% nDCG@10 using relevant feedback documents.

Ramraj Chandradevan, Kaustubh D. Dhole, Eugene Agichtein · amazon-science

State-of-the-art neural rankers pre-trained on large task-specific training data such as MS-MARCO, have been shown to exhibit strong performance on various ranking tasks without domain adaptation, also called zero-shot. However, zero-shot neural ranking may be sub-optimal, as it does not take advantage of the target domain information. Unfortunately, acquiring sufficiently large and high quality target training data to improve a modern neural ranker can be costly and time-consuming. To address this problem, we propose a new approach to unsupervised domain adaptation for ranking, DUQGen, which addresses a critical gap in prior literature, namely how to automatically generate both effective and diverse synthetic training data to fine tune a modern neural ranker for a new domain. Specifically, DUQGen produces a more effective representation of the target domain by identifying clusters of similar documents; and generates a more diverse training dataset by probabilistic sampling over the resulting document clusters. Our extensive experiments, over the standard BEIR collection, demonstrate that DUQGen consistently outperforms all zero-shot baselines and substantially outperforms the SOTA baselines on 16 out of 18 datasets, for an average of 4% relative improvement across all datasets. We complement our results with a thorough analysis for more in-depth understanding of the proposed method's performance and to identify promising areas for further improvements.

Kaustubh D. Dhole, Kai Shu, Eugene Agichtein

Computational argumentation, which involves generating answers or summaries for controversial topics like abortion bans and vaccination, has become increasingly important in today's polarized environment. Sophisticated LLM capabilities offer the potential to provide nuanced, evidence-based answers to such questions through Retrieval-Augmented Argumentation (RAArg), leveraging real-world evidence for high-quality, grounded arguments. However, evaluating RAArg remains challenging, as human evaluation is costly and difficult for complex, lengthy answers on complicated topics. At the same time, re-using existing argumentation datasets is no longer sufficient, as they lack long, complex arguments and realistic evidence from potentially misleading sources, limiting holistic evaluation of retrieval effectiveness and argument quality. To address these gaps, we investigate automated evaluation methods using multiple fine-grained LLM judges, providing better and more interpretable assessments than traditional single-score metrics and even previously reported human crowdsourcing. To validate the proposed techniques, we introduce ConQRet, a new benchmark featuring long and complex human-authored arguments on debated topics, grounded in real-world websites, allowing an exhaustive evaluation across retrieval effectiveness, argument quality, and groundedness. We validate our LLM Judges on a prior dataset and the new ConQRet benchmark. Our proposed LLM Judges and the ConQRet benchmark can enable rapid progress in computational argumentation and can be naturally extended to other complex retrieval-augmented generation tasks.

Kaustubh D. Dhole, Nikhita Vedula, Saar Kuzi et al. · amazon-science

In Recommender Systems, users often seek the best products through indirect, vague, or under-specified queries, such as "best shoes for trail running". Such queries, also referred to as implicit superlative queries, pose a significant challenge for standard retrieval and ranking systems as they lack an explicit mention of attributes and require identifying and reasoning over complex factors. We investigate how Large Language Models (LLMs) can generate implicit attributes for ranking as well as reason over them to improve product recommendations for such queries. As a first step, we propose a novel four-point schema for annotating the best product candidates for superlative queries called SUPERB, paired with LLM-based product annotations. We then empirically evaluate several existing retrieval and ranking approaches on our new dataset, providing insights and discussing their integration into real-world e-commerce production systems.

Chase Fensore, Kaustubh Dhole, Joyce C Ho et al.

We present our submission to the LiveRAG Challenge 2025, which evaluates retrieval-augmented generation (RAG) systems on dynamic test sets using the FineWeb-10BT corpus. Our final hybrid approach combines sparse (BM25) and dense (E5) retrieval methods and then aims to generate relevant and faithful answers with Falcon3-10B-Instruct. Through systematic evaluation on 200 synthetic questions generated with DataMorgana across 64 unique question-user combinations, we demonstrate that neural re-ranking with RankLLaMA improves MAP from 0.523 to 0.797 (52% relative improvement) but introduces prohibitive computational costs (84s vs 1.74s per question). While DSPy-optimized prompting strategies achieved higher semantic similarity (0.771 vs 0.668), their 0% refusal rates raised concerns about over-confidence and generalizability. Our submitted hybrid system without re-ranking achieved 4th place in faithfulness and 11th place in correctness among 25 teams. Analysis across question categories reveals that vocabulary alignment between questions and documents was the strongest predictor of performance on our development set, with document-similar phrasing improving cosine similarity from 0.562 to 0.762.

Nikhita Vedula, Giuseppe Castellucci, Eugene Agichtein et al. · amazon-science

Conversational Task Assistants (CTAs) guide users in performing a multitude of activities, such as making recipes. However, ensuring that interactions remain engaging, interesting, and enjoyable for CTA users is not trivial, especially for time-consuming or challenging tasks. Grounded in psychological theories of human interest, we propose to engage users with contextual and interesting statements or facts during interactions with a multi-modal CTA, to reduce fatigue and task abandonment before a task is complete. To operationalize this idea, we train a high-performing classifier (82% F1-score) to automatically identify relevant and interesting facts for users. We use it to create an annotated dataset of task-specific interesting facts for the domain of cooking. Finally, we design and validate a dialogue policy to incorporate the identified relevant and interesting facts into a conversation, to improve user engagement and task completion. Live testing on a leading multi-modal voice assistant shows that 66% of the presented facts were received positively, leading to a 40% gain in the user satisfaction rating, and a 37% increase in conversation length. These findings emphasize that strategically incorporating interesting facts into the CTA experience can promote real-world user participation for guided task interactions.

Jianghong Zhou, Eugene Agichtein

To support complex search tasks, where the initial information requirements are complex or may change during the search, a search engine must adapt the information delivery as the user's information requirements evolve. To support this dynamic ranking paradigm effectively, search result ranking must incorporate both the user feedback received, and the information displayed so far. To address this problem, we introduce a novel reinforcement learning-based approach, RLIrank. We first build an adapted reinforcement learning framework to integrate the key components of the dynamic search. Then, we implement a new Learning to Rank (LTR) model for each iteration of the dynamic search, using a recurrent Long Short Term Memory neural network (LSTM), which estimates the gain for each next result, learning from each previously ranked document. To incorporate the user's feedback, we develop a word-embedding variation of the classic Rocchio Algorithm, to help guide the ranking towards the high-value documents. Those innovations enable RLIrank to outperform the previously reported methods from the TREC Dynamic Domain Tracks 2017 and exceed all the methods in 2016 TREC Dynamic Domain after multiple search iterations, advancing the state of the art for dynamic search.

Jianghong Zhou, Eugene Agichtein, Surya Kallumadi

In search and recommendation, diversifying the multi-aspect search results could help with reducing redundancy, and promoting results that might not be shown otherwise. Many previous methods have been proposed for this task. However, previous methods do not explicitly consider the uniformity of the number of the items' classes, or evenness, which could degrade the search and recommendation quality. To address this problem, we introduce a novel method by adapting the Simpson's Diversity Index from biology, which enables a more effective and efficient quadratic search result diversification algorithm. We also extend the method to balance the diversity between multiple aspects through weighted factors and further improve computational complexity by developing a fast approximation algorithm. We demonstrate the feasibility of the proposed method using the openly available Kaggle shoes competition dataset. Our experimental results show that our approach outperforms previous state of the art diversification methods, while reducing computational complexity.

Jianghong Zhou, Sayyed M. Zahiri, Simon Hughes et al.

Users' clicks on Web search results are one of the key signals for evaluating and improving web search quality and have been widely used as part of current state-of-the-art Learning-To-Rank(LTR) models. With a large volume of search logs available for major search engines, effective models of searcher click behavior have emerged to evaluate and train LTR models. However, when modeling the users' click behavior, considering the bias of the behavior is imperative. In particular, when a search result is not clicked, it is not necessarily chosen as not relevant by the user, but instead could have been simply missed, especially for lower-ranked results. These kinds of biases in the click log data can be incorporated into the click models, propagating the errors to the resulting LTR ranking models or evaluation metrics. In this paper, we propose the De-biased Reinforcement Learning Click model (DRLC). The DRLC model relaxes previously made assumptions about the users' examination behavior and resulting latent states. To implement the DRLC model, convolutional neural networks are used as the value networks for reinforcement learning, trained to learn a policy to reduce bias in the click logs. To demonstrate the effectiveness of the DRLC model, we first compare performance with the previous state-of-art approaches using established click prediction metrics, including log-likelihood and perplexity. We further show that DRLC also leads to improvements in ranking performance. Our experiments demonstrate the effectiveness of the DRLC model in learning to reduce bias in click logs, leading to improved modeling performance and showing the potential for using DRLC for improving Web search quality.

Ali Ahmadvand, Surya Kallumadi, Faizan Javed et al.

Mapping a search query to a set of relevant categories in the product taxonomy is a significant challenge in e-commerce search for two reasons: 1) Training data exhibits severe class imbalance problem due to biased click behavior, and 2) queries with little customer feedback (e.g., tail queries) are not well-represented in the training set, and cause difficulties for query understanding. To address these problems, we propose a deep learning model, DeepCAT, which learns joint word-category representations to enhance the query understanding process. We believe learning category interactions helps to improve the performance of category mapping on minority classes, tail and torso queries. DeepCAT contains a novel word-category representation model that trains the category representations based on word-category co-occurrences in the training set. The category representation is then leveraged to introduce a new loss function to estimate the category-category co-occurrences for refining joint word-category embeddings. To demonstrate our model's effectiveness on minority categories and tail queries, we conduct two sets of experiments. The results show that DeepCAT reaches a 10% improvement on minority classes and a 7.1% improvement on tail queries over a state-of-the-art label embedding model. Our findings suggest a promising direction for improving e-commerce search by semantic modeling of taxonomy hierarchies.

Ali Ahmadvand, Sayyed M. Zahiri, Simon Hughes et al.

Query categorization is an essential part of query intent understanding in e-commerce search. A common query categorization task is to select the relevant fine-grained product categories in a product taxonomy. For frequent queries, rich customer behavior (e.g., click-through data) can be used to infer the relevant product categories. However, for more rare queries, which cover a large volume of search traffic, relying solely on customer behavior may not suffice due to the lack of this signal. To improve categorization of rare queries, we adapt the Pseudo-Relevance Feedback (PRF) approach to utilize the latent knowledge embedded in semantically or lexically similar product documents to enrich the representation of the more rare queries. To this end, we propose a novel deep neural model named Attentive Pseudo Relevance Feedback Network (APRF-Net) to enhance the representation of rare queries for query categorization. To demonstrate the effectiveness of our approach, we collect search queries from a large commercial search engine, and compare APRF-Net to state-of-the-art deep learning models for text classification. Our results show that the APRF-Net significantly improves query categorization by 5.9% on F1@1 score over the baselines, which increases to 8.2% improvement for the rare (tail) queries. The findings of this paper can be leveraged for further improvements in search query representation and understanding.

Iftah Gamzu, Hila Gonen, Gilad Kutiel et al.

In recent years online shopping has gained momentum and became an important venue for customers wishing to save time and simplify their shopping process. A key advantage of shopping online is the ability to read what other customers are saying about products of interest. In this work, we aim to maintain this advantage in situations where extreme brevity is needed, for example, when shopping by voice. We suggest a novel task of extracting a single representative helpful sentence from a set of reviews for a given product. The selected sentence should meet two conditions: first, it should be helpful for a purchase decision and second, the opinion it expresses should be supported by multiple reviewers. This task is closely related to the task of Multi Document Summarization in the product reviews domain but differs in its objective and its level of conciseness. We collect a dataset in English of sentence helpfulness scores via crowd-sourcing and demonstrate its reliability despite the inherent subjectivity involved. Next, we describe a complete model that extracts representative helpful sentences with positive and negative sentiment towards the product and demonstrate that it outperforms several baselines.

Jason Ingyu Choi, Surya Kallumadi, Bhaskar Mitra et al.

Retrieving all semantically relevant products from the product catalog is an important problem in E-commerce. Compared to web documents, product catalogs are more structured and sparse due to multi-instance fields that encode heterogeneous aspects of products (e.g. brand name and product dimensions). In this paper, we propose a new semantic product search algorithm that learns to represent and aggregate multi-instance fields into a document representation using state of the art transformers as encoders. Our experiments investigate two aspects of the proposed approach: (1) effectiveness of field representations and structured matching; (2) effectiveness of adding lexical features to semantic search. After training our models using user click logs from a well-known E-commerce platform, we show that our results provide useful insights for improving product search. Lastly, we present a detailed error analysis to show which types of queries benefited the most by fielded representations and structured matching.

Harshita Sahijwani, Jason Ingyu Choi, Eugene Agichtein

One of the key benefits of voice-based personal assistants is the potential to proactively recommend relevant and interesting information. One of the most valuable sources of such information is the News. However, in order for the user to hear the news that is useful and relevant to them, it must be recommended in an interesting and informative way. However, to the best of our knowledge, how to present a news item for a voice-based recommendation remains an open question. In this paper, we empirically compare different ways of recommending news, or specific news items, in a voice-based conversational setting. Specifically, we study the user engagement and satisfaction with five different variants of presenting news recommendations: (1) a generic news briefing; (2) news about a specific entity relevant to the current conversation; (3) news about an entity from a past conversation; (4) news on a trending news topic; and (5) the default - a suggestion to talk about news in general. Our results show that entity-based news recommendations exhibit 29% higher acceptance compared to briefing recommendations, and almost 100% higher acceptance compared to recommending generic or trending news. Our investigation into the presentation of news recommendations and the resulting insights could make voice assistants more informative and engaging.

Jason Ingyu Choi, Ali Ahmadvand, Eugene Agichtein

Predicting user satisfaction in conversational systems has become critical, as spoken conversational assistants operate in increasingly complex domains. Online satisfaction prediction (i.e., predicting satisfaction of the user with the system after each turn) could be used as a new proxy for implicit user feedback, and offers promising opportunities to create more responsive and effective conversational agents, which adapt to the user's engagement with the agent. To accomplish this goal, we propose a conversational satisfaction prediction model specifically designed for open-domain spoken conversational agents, called ConvSAT. To operate robustly across domains, ConvSAT aggregates multiple representations of the conversation, namely the conversation history, utterance and response content, and system- and user-oriented behavioral signals. We first calibrate ConvSAT performance against state of the art methods on a standard dataset (Dialogue Breakdown Detection Challenge) in an online regime, and then evaluate ConvSAT on a large dataset of conversations with real users, collected as part of the Alexa Prize competition. Our experimental results show that ConvSAT significantly improves satisfaction prediction for both offline and online setting on both datasets, compared to the previously reported state-of-the-art approaches. The insights from our study can enable more intelligent conversational systems, which could adapt in real-time to the inferred user satisfaction and engagement.

Jason Ingyu Choi, Eugene Agichtein

As voice-based assistants such as Alexa, Siri, and Google Assistant become ubiquitous, users increasingly expect to maintain natural and informative conversations with such systems. However, for an open-domain conversational system to be coherent and engaging, it must be able to maintain the user's interest for extended periods, without sounding boring or annoying. In this paper, we investigate one natural approach to this problem, of modulating response prosody, i.e., changing the pitch and cadence of the response to indicate delight, sadness or other common emotions, as well as using pre-recorded interjections. Intuitively, this approach should improve the naturalness of the conversation, but attempts to quantify the effects of prosodic modulation on user satisfaction and engagement remain challenging. To accomplish this, we report results obtained from a large-scale empirical study that measures the effects of prosodic modulation on user behavior and engagement across multiple conversation domains, both immediately after each turn, and at the overall conversation level. Our results indicate that the prosody modulation significantly increases both immediate and overall user satisfaction. However, since the effects vary across different domains, we verify that prosody modulations do not substitute for coherent, informative content of the responses. Together, our results provide useful tools and insights for improving the naturalness of responses in conversational systems.

Ali Ahmadvand, Jason Ingyu Choi, Eugene Agichtein

Classifying the general intent of the user utterance in a conversation, also known as Dialogue Act (DA), e.g., open-ended question, statement of opinion, or request for an opinion, is a key step in Natural Language Understanding (NLU) for conversational agents. While DA classification has been extensively studied in human-human conversations, it has not been sufficiently explored for the emerging open-domain automated conversational agents. Moreover, despite significant advances in utterance-level DA classification, full understanding of dialogue utterances requires conversational context. Another challenge is the lack of available labeled data for open-domain human-machine conversations. To address these problems, we propose a novel method, CDAC (Contextual Dialogue Act Classifier), a simple yet effective deep learning approach for contextual dialogue act classification. Specifically, we use transfer learning to adapt models trained on human-human conversations to predict dialogue acts in human-machine dialogues. To investigate the effectiveness of our method, we train our model on the well-known Switchboard human-human dialogue dataset, and fine-tune it for predicting dialogue acts in human-machine conversation data, collected as part of the Amazon Alexa Prize 2018 competition. The results show that the CDAC model outperforms an utterance-level state of the art baseline by 8.0% on the Switchboard dataset, and is comparable to the latest reported state-of-the-art contextual DA classification results. Furthermore, our results show that fine-tuning the CDAC model on a small sample of manually labeled human-machine conversations allows CDAC to more accurately predict dialogue acts in real users' conversations, suggesting a promising direction for future improvements.

Ali Ahmadvand, Harshita Sahijwani, Eugene Agichtein

To hold a true conversation, an intelligent agent should be able to occasionally take initiative and recommend the next natural conversation topic. This is a challenging task. A topic suggested by the agent should be relevant to the person, appropriate for the conversation context, and the agent should have something interesting to say about it. Thus, a scripted, or one-size-fits-all, popularity-based topic suggestion is doomed to fail. Instead, we explore different methods for a personalized, contextual topic suggestion for open-domain conversations. We formalize the Conversational Topic Suggestion problem (CTS) to more clearly identify the assumptions and requirements. We also explore three possible approaches to solve this problem: (1) model-based sequential topic suggestion to capture the conversation context (CTS-Seq), (2) Collaborative Filtering-based suggestion to capture previous successful conversations from similar users (CTS-CF), and (3) a hybrid approach combining both conversation context and collaborative filtering. To evaluate the effectiveness of these methods, we use real conversations collected as part of the Amazon Alexa Prize 2018 Conversational AI challenge. The results are promising: the CTS-Seq model suggests topics with 23% higher accuracy than the baseline, and incorporating collaborative filtering signals into a hybrid CTS-Seq-CF model further improves recommendation accuracy by 12%. Together, our proposed models, experiments, and analysis significantly advance the study of open-domain conversational agents, and suggest promising directions for future improvements.

Ali Ahmadvand, Harshita Sahijwani, Jason Ingyu Choi et al.

Identifying the topic (domain) of each user's utterance in open-domain conversational systems is a crucial step for all subsequent language understanding and response tasks. In particular, for complex domains, an utterance is often routed to a single component responsible for that domain. Thus, correctly mapping a user utterance to the right domain is critical. To address this problem, we introduce ConCET: a Concurrent Entity-aware conversational Topic classifier, which incorporates entity-type information together with the utterance content features. Specifically, ConCET utilizes entity information to enrich the utterance representation, combining character, word, and entity-type embeddings into a single representation. However, for rich domains with millions of available entities, unrealistic amounts of labeled training data would be required. To complement our model, we propose a simple and effective method for generating synthetic training data, to augment the typically limited amounts of labeled training data, using commonly available knowledge bases to generate additional labeled utterances. We extensively evaluate ConCET and our proposed training method first on an openly available human-human conversational dataset called Self-Dialogue, to calibrate our approach against previous state-of-the-art methods; second, we evaluate ConCET on a large dataset of human-machine conversations with real users, collected as part of the Amazon Alexa Prize. Our results show that ConCET significantly improves topic classification performance on both datasets, including 8-10% improvements over state-of-the-art deep learning methods. We complement our quantitative results with detailed analysis of system performance, which could be used for further improvements of conversational agents.

Ali Ahmadvand, Surya Kallumadi, Faizan Javed et al.

An accurate understanding of a user's query intent can help improve the performance of downstream tasks such as query scoping and ranking. In the e-commerce domain, recent work in query understanding focuses on the query to product-category mapping. But, a small yet significant percentage of queries (in our website 1.5% or 33M queries in 2019) have non-commercial intent associated with them. These intents are usually associated with non-commercial information seeking needs such as discounts, store hours, installation guides, etc. In this paper, we introduce Joint Query Intent Understanding (JointMap), a deep learning model to simultaneously learn two different high-level user intent tasks: 1) identifying a query's commercial vs. non-commercial intent, and 2) associating a set of relevant product categories in taxonomy to a product query. JointMap model works by leveraging the transfer bias that exists between these two related tasks through a joint-learning process. As curating a labeled data set for these tasks can be expensive and time-consuming, we propose a distant supervision approach in conjunction with an active learning model to generate high-quality training data sets. To demonstrate the effectiveness of JointMap, we use search queries collected from a large commercial website. Our results show that JointMap significantly improves both "commercial vs. non-commercial" intent prediction and product category mapping by 2.3% and 10% on average over state-of-the-art deep learning methods. Our findings suggest a promising direction to model the intent hierarchies in an e-commerce search engine.

Payam Karisani, Joyce C. Ho, Eugene Agichtein

Mining social media content for tasks such as detecting personal experiences or events, suffer from lexical sparsity, insufficient training data, and inventive lexicons. To reduce the burden of creating extensive labeled data and improve classification performance, we propose to perform these tasks in two steps: 1. Decomposing the task into domain-specific sub-tasks by identifying key concepts, thus utilizing human domain understanding; and 2. Combining the results of learners for each key concept using co-training to reduce the requirements for labeled training data. We empirically show the effectiveness and generality of our approach, Co-Decomp, using three representative social media mining tasks, namely Personal Health Mention detection, Crisis Report detection, and Adverse Drug Reaction monitoring. The experiments show that our model is able to outperform the state-of-the-art text classification models--including those using the recently introduced BERT model--when small amounts of training data are available.

Dana Sagi, Tzoof Avny, Kira Radinsky et al.

One of the main challenges in ranking is embedding the query and document pairs into a joint feature space, which can then be fed to a learning-to-rank algorithm. To achieve this representation, the conventional state of the art approaches perform extensive feature engineering that encode the similarity of the query-answer pair. Recently, deep-learning solutions have shown that it is possible to achieve comparable performance, in some settings, by learning the similarity representation directly from data. Unfortunately, previous models perform poorly on longer texts, or on texts with significant portion of irrelevant information, or which are grammatically incorrect. To overcome these limitations, we propose a novel ranking algorithm for question answering, QARAT, which uses an attention mechanism to learn on which words and phrases to focus when building the mutual representation. We demonstrate superior ranking performance on several real-world question-answer ranking datasets, and provide visualization of the attention mechanism to otter more insights into how our models of attention could benefit ranking for difficult question answering challenges.

Payam Karisani, Eugene Agichtein

Millions of users share their experiences on social media sites, such as Twitter, which in turn generate valuable data for public health monitoring, digital epidemiology, and other analyses of population health at global scale. The first, critical, task for these applications is classifying whether a personal health event was mentioned, which we call the (PHM) problem. This task is challenging for many reasons, including typically short length of social media posts, inventive spelling and lexicons, and figurative language, including hyperbole using diseases like "heart attack" or "cancer" for emphasis, and not as a health self-report. This problem is even more challenging for rarely reported, or frequent but ambiguously expressed conditions, such as "stroke". To address this problem, we propose a general, robust method for detecting PHMs in social media, which we call WESPAD, that combines lexical, syntactic, word embedding-based, and context-based features. WESPAD is able to generalize from few examples by automatically distorting the word embedding space to most effectively detect the true health mentions. Unlike previously proposed state-of-the-art supervised and deep-learning techniques, WESPAD requires relatively little training data, which makes it possible to adapt, with minimal effort, to each new disease and condition. We evaluate WESPAD on both an established publicly available Flu detection benchmark, and on a new dataset that we have constructed with mentions of multiple health conditions. Our experiments show that WESPAD outperforms the baselines and state-of-the-art methods, especially in cases when the number and proportion of true health mentions in the training data is small.