The page presentation biases in the information retrieval system, especially on the click behavior, is a well-known challenge that hinders improving ranking models' performance with implicit user feedback. Unbiased Learning to Rank~(ULTR) algorithms are then proposed to learn an unbiased ranking model with biased click data. However, most existing algorithms are specifically designed to mitigate position-related bias, e.g., trust bias, without considering biases induced by other features in search result page presentation(SERP), e.g. attractive bias induced by the multimedia. Unfortunately, those biases widely exist in industrial systems and may lead to an unsatisfactory search experience. Therefore, we introduce a new problem, i.e., whole-page Unbiased Learning to Rank(WP-ULTR), aiming to handle biases induced by whole-page SERP features simultaneously. It presents tremendous challenges: (1) a suitable user behavior model (user behavior hypothesis) can be hard to find; and (2) complex biases cannot be handled by existing algorithms. To address the above challenges, we propose a Bias Agnostic whole-page unbiased Learning to rank algorithm, named BAL, to automatically find the user behavior model with causal discovery and mitigate the biases induced by multiple SERP features with no specific design. Experimental results on a real-world dataset verify the effectiveness of the BAL.
论文检索
输入标题、作者或关键词，从 432 篇学术成果中精准定位
The similarity matrix is at the core of similarity search problems. However, incomplete observations are ubiquitous in real scenarios leading to a less accurate similarity matrix. To alleviate this problem, in this paper, based on the key insight that the similarity matrix enjoys both the symmetric and positive semi-definiteness (PSD) properties, we propose a novel similarity matrix calibration method, which is scalable, effective, and sound. Specifically, we establish the PSD property as a constraint for the similarity matrix calibration problem and propose a novel similarity matrix calibration method to estimate the similarity matrix, which approximates the unknown complete ground-truth similarity matrix. To enable a fast optimization process, we further develop a general approximated algorithm that bypasses the computation of singular values. Theoretical analysis ensures stable calibration performance and convergence speed. Extensive experiments of similarity matrix calibration on real-world datasets demonstrate that our proposed method outperforms baseline methods in terms of both accuracy and speed.
Conversational search has seen increased recent attention in both the IR and NLP communities. It seeks to clarify and solve users' search needs through multi-turn natural language interactions. However, most existing systems are trained and demonstrated with recorded or artificial conversation logs. Eventually, conversational search systems should be trained, evaluated, and deployed in an open-ended setting with unseen conversation trajectories. A key challenge is that training and evaluating such systems both require a human-in-the-loop, which is expensive and does not scale. One strategy is to simulate users, thereby reducing the scaling costs. However, current user simulators are either limited to only responding to yes-no questions from the conversational search system or unable to produce high-quality responses in general. This paper shows that existing user simulation systems could be significantly improved by a smaller finetuned natural language generation model. However, rather than merely reporting it as the new state-of-the-art, we consider it a strong baseline and present an in-depth investigation of simulating user response for conversational search. Our goal is to supplement existing work with an insightful hand-analysis of unsolved challenges by the baseline and propose our solutions. The challenges we identified include (1) a blind spot that is difficult to learn, and (2) a specific type of misevaluation in the standard setup. We propose a new generation system to effectively cover the training blind spot and suggest a new evaluation setup to avoid misevaluation. Our proposed system leads to significant improvements over existing systems and large language models such as GPT-4. Additionally, our analysis provides insights into the nature of the task to facilitate future work.
Unsupervised semantic hashing has emerged as an indispensable technique for fast image search, which aims to convert images into binary hash codes without relying on labels. Recent advancements in the field demonstrate that employing large-scale backbones (e.g., ViT) in unsupervised semantic hashing models can yield substantial improvements. However, the inference delay has become increasingly difficult to overlook. Knowledge distillation provides a means for practical model compression to alleviate this delay. Nevertheless, the prevailing knowledge distillation approaches are not explicitly designed for semantic hashing. They ignore the unique search paradigm of semantic hashing, the inherent necessities of the distillation process, and the property of hash codes. In this paper, we propose an innovative Bit-mask Robust Contrastive knowledge Distillation (BRCD) method, specifically devised for the distillation of semantic hashing models. To ensure the effectiveness of two kinds of search paradigms in the context of semantic hashing, BRCD first aligns the semantic spaces between the teacher and student models through a contrastive knowledge distillation objective. Additionally, to eliminate noisy augmentations and ensure robust optimization, a cluster-based method within the knowledge distillation process is introduced. Furthermore, through a bit-level analysis, we uncover the presence of redundancy bits resulting from the bit independence property. To mitigate these effects, we introduce a bit mask mechanism in our knowledge distillation objective. Finally, extensive experiments not only showcase the noteworthy performance of our BRCD method in comparison to other knowledge distillation methods but also substantiate the generality of our methods across diverse semantic hashing models and backbones. The code for BRCD is available at https://github.com/hly1998/BRCD.
Users' queries are usually vague, and their search intents tend to be ambiguous, thereby needing search clarification to clarify users' current intent by asking a clarifying question and providing several clickable sub-intent items as clarification options. However, in addition to drilling down the current query, users may also have exploratory needs that diverge from their current intent. For example, a user searching for the query "Cartier women watches'' may also potentially want to explore some parallel information by issuing queries such as "Rolex women watches'' or "Cartier women bracelets'', named exploratory queries in this paper. These exploratory needs are common during the search process yet cannot be satisfied by current search clarification approaches which typically stick to the sub-intents of the query. This paper focuses on mining exploratory queries as additional options to meet users' exploratory needs in conversational search systems. Specifically, we first design a rule-based model that generates exploratory queries based on the current query's top retrieved documents. Then, we propose using the data generated by the rule-based model to train a neural generation model through multi-task learning for further generalization. Finally, we borrow the in-context learning ability of the large language model to generate exploratory queries based on prompt engineering. We constructed an evaluation dataset based on human annotations and conduct an extensive set of experiments. The results show that our proposed methods generate higher-quality exploratory queries compared with several baselines.
In the rapidly evolving landscape of information retrieval, search engines strive to provide more personalized and relevant results to users. Query suggestion systems play a crucial role in achieving this goal by assisting users in formulating effective queries. However, existing query suggestion systems mainly rely on textual inputs, potentially limiting user search experiences for querying images. In this paper, we introduce a novel Multimodal Query Suggestion (MMQS) task, which aims to generate query suggestions based on user query images to improve the intentionality and diversity of search results. We present the RL4Sugg framework, leveraging the power of Large Language Models (LLMs) with Multi-Agent Reinforcement Learning from Human Feedback to optimize the generation process. Through comprehensive experiments, we validate the effectiveness of RL4Sugg, demonstrating a 18% improvement compared to the best existing approach. Moreover, the MMQS has been transferred into real-world search engine products, which yield enhanced user engagement. Our research advances query suggestion systems and provides a new perspective on multimodal information retrieval.
Making the contents generated by Large Language Model (LLM), accurate, credible and traceable is crucial, especially in complex knowledge-intensive tasks that require multi-step reasoning and each step needs knowledge to solve. Retrieval-augmented generation is good potential to solve this problem. However, where and how to introduce Information Retrieval (IR) to LLM is a big challenge. Previous work has the problems that wrong knowledge retrieved by IR misleads the LLM and interaction between IR and LLM breaks the reasoning chain of LLM. This paper proposes a novel framework named Search-in-the-Chain (SearChain) for the interaction between LLM and IR to solve the challenges. First, LLM generates the reasoning chain named Chain-of-Query (CoQ) where each node consists of an IR-oriented query-answer pair. Second, IR verifies the answer of each node of CoQ. It corrects the answer that is not consistent with the retrieved information when IR gives high confidence, which improves the credibility. Third, LLM can indicate its missing knowledge in CoQ and rely on IR to provide this knowledge to LLM. These operations improve the accuracy in terms of reasoning and knowledge. Finally, SearChain generates the reasoning process and marks references to supporting documents for each reasoning step, which improves traceability. Interaction with IR in SearChain forms a novel reasoning path based on a tree, which enables LLM to dynamically modify the direction of reasoning. Experiments show that SearChain outperforms state-of-the-art baselines on complex knowledge-intensive tasks including multi-hop Q&A, slot filling, fact checking, and long-form Q&A.
Ranking items regarding individual user interests is a core technique of multiple downstream tasks such as recommender systems. Learning such a personalized ranker typically relies on the implicit feedback from users' past click-through behaviors. However, collected feedback is biased toward previously highly-ranked items and directly learning from it would result in "rich-get-richer" phenomena. In this paper, we propose a simple yet sufficient unbiased learning-to-rank paradigm named InfoRank that aims to simultaneously address both position and popularity biases. We begin by consolidating the impacts of those biases into a single observation factor, thereby providing a unified approach to addressing bias-related issues. Subsequently, we minimize the mutual information between the observation estimation and the relevance estimation conditioned on the input features. By doing so, our relevance estimation can be proved to be free of bias. To implement InfoRank, we first incorporate an attention mechanism to capture latent correlations within user-item features, thereby generating estimations of observation and relevance. We then introduce a regularization term, grounded in conditional mutual information, to promote conditional independence between relevance estimation and observation estimation. Experimental evaluations conducted across three extensive recommendation and search datasets reveal that InfoRank learns more precise and unbiased ranking strategies.
In the search process, it is essential to strike a balance between effectiveness and efficiency to improve search experience. Thus, ranking list truncation has become increasingly crucial. Especially in the legal domain, irrelevant cases can severely increase search costs and even compromise the pursuit of legal justice. However, there are truncation challenges that mainly arise from the distinctive structure of legal case documents, where the elements such as fact, reasoning, and judgement in a case serve as different but multi-view texts, which could result in a bad performance if the multi-view texts cannot be well-modeled. Existing approaches are limited due to their inability to handle multi-view elements information and their neglect of semantic interconnections between cases in the ranking list. In this paper, we propose a multi-view truncation framework for legal case retrieval, named MileCut. MileCut employs a case elements extraction module to fully exploit the multi-view information of cases in the ranking list. Then, MileCut applies a multi-view truncation module to select the most informative view and make a more comprehensive cut-off decision, similar to how legal experts look over retrieval results. As a practical evaluation, MileCut is assessed across three datasets, including criminal and civil case retrieval scenarios, and the results show that MileCut outperforms other methods on F1, DCG, and OIE metrics.
The results of information retrieval (IR) are usually presented in the form of a ranking list of candidate documents, such as web search for humans and retrieval-augmented generation for large language models (LLMs). List-aware retrieval aims to capture the list-level contextual features to return a better list, mainly including reranking and truncation. Reranking finely re-scores the documents in the list. Truncation dynamically determines the cut-off point of the ranked list to achieve the trade-off between overall relevance and avoiding misinformation from irrelevant documents. Previous studies treat them as two separate tasks and model them separately. However, the separation is not optimal. First, it is hard to share the contextual information of the ranking list between the two tasks. Second, the separate pipeline usually meets the error accumulation problem, where the small error from the reranking stage can largely affect the truncation stage. To solve these problems, we propose a Reranking-Truncation joint model (GenRT) that can perform the two tasks concurrently. GenRT integrates reranking and truncation via a generative paradigm based on an encoder-decoder architecture with novel loss functions for joint optimization to learn both tasks. Sharing parameters by the joint model is conducive to making full use of the common modeling information of the two tasks. Besides, the two tasks are performed concurrently and co-optimized to solve the error accumulation problem between separate stages. Experiments on public learning-to-rank benchmarks and open-domain Q&A tasks show that our method achieves SOTA performance on both reranking and truncation tasks for web search and retrieval-augmented LLMs.
When making recommendations, there is an apparent trade-off between the goals of accuracy (to recommend items a user is most likely to want) and diversity (to recommend items representing a range of categories). As such, real-world recommender systems often explicitly incorporate diversity into recommendations, at the cost of accuracy. We study the accuracy-diversity trade-off by bringing in a third concept: user utility. We argue that accuracy is misaligned with user utility because it fails to incorporate a user's consumption constraints: at any given time, users can typically only use at most a few recommended items (e.g., dine at one restaurant, or watch a couple of movies). In a theoretical model, we show that utility-maximizing recommendations---when accounting for consumption constraints---are naturally diverse due to diminishing returns of recommending similar items. Therefore, while increasing diversity may come at the cost of accuracy, it can also help align accuracy-based recommendations toward the more fundamental objective of user utility. Our theoretical results yield practical guidance into how recommendations should incorporate diversity to serve user ends.
We study the impact of content moderation policies in online communities. In our theoretical model, a platform chooses a content moderation policy and individuals choose whether or not to participate in the community according to the fraction of user content that aligns with their preferences. The effects of content moderation, at first blush, might seem obvious: platform speech is restricted. However, when user participation decisions are taken into account, its effects can be more subtle --- and counter-intuitive. For example, our model can straightforwardly demonstrate how moderation policies mayincrease participation and/ordiversify content available on the platform. In our analysis, we explore a rich set of interconnected phenomena related to content moderation in online communities. We first characterize the effectiveness of a natural class of moderation policies for creating and sustaining communities. Building on this, we explore how resource-limited or ideological platforms might set policies, how communities are affected by differing levels of personalization, and competition between platforms. Our model provides a vocabulary and mathematically tractable framework for analyzing platform decisions about content moderation.
Investigating how websites use sensitive user data is an active research area. However, research based on automated measurements has been limited to those websites that do not require user authentication. To overcome this limitation, we developed a crawler that automates website registrations and newsletter subscriptions and detects both security and privacy threats at scale. We demonstrate our crawler's capabilities by running it on 660k websites. We use this to identify security and privacy threats and to contextualize them within EU laws, namely the General Data Protection Regulation and ePrivacy Directive. Our methods detect private data collection over insecure HTTP connections and websites sending emails with user-provided passwords. We are also the first to apply machine learning to web forms, assessing violations of marketing consent collection requirements. Overall, we find that 37.2% of websites send marketing emails without proper user consent. This is mostly caused by websites failing both to verify and store consent adequately. Additionally, 1.8% of websites share users' email addresses with third parties without a transparent disclosure.
Browser extensions offer a variety of valuable features and functionalities. They also pose a significant security risk if not properly designed or reviewed. Prior works have shown that browser extensions can access and manipulate data fields, including sensitive data such as passwords, credit card numbers, and Social Security numbers. In this paper, we present an empirical study of the security risks posed by browser extensions. Specifically, we first build a proof-of-concept extension that can steal sensitive user information. We find that the extension passes the Chrome Webstore review process. We then perform a measurement study on the top 10K website login pages to check if the extension access to password fields via JS. We find that none of the password fields are actively protected, and can be accessed using JS. Moreover, we found that 1K websites store passwords in plaintext in their page source, including popular websites like Google.com and Cloudflare.com. We also analyzed over 160K Chrome Web Store extensions for malicious behavior, finding that 28K have permission to access sensitive fields and 190 store password fields in variables. To analyze the behavioral workflow of the potentially malicious extensions, we propose an LLM-driven framework, Extension Reviewer. Finally, we discuss two countermeasures to address these risks: a bolt-on JavaScript package for immediate adoption by website developers allowing them to protect sensitive input fields, and a browser-level solution that alerts users when an extension accesses sensitive input fields. Our research highlights the urgent need for improved security measures to protect sensitive user information online.
Fake news is pervasive on social media, inflicting substantial harm on public discourse and societal well-being. We investigate the explicit structural information and textual features of news pieces by constructing a heterogeneous graph concerning the relations among news topics, entities, and content. Through our study, we reveal that fake news can be effectively detected in terms of the atypical heterogeneous subgraphs centered on them, which encapsulate the essential semantics and intricate relations between news elements. However, suffering from the heterogeneity, exploring such heterogeneous subgraphs remains an open problem. To bridge the gap, this work proposes a heterogeneous subgraph transformer HeteroSGT to exploit subgraphs in our constructed heterogeneous graph. In HeteroSGT, we first employ a pre-trained language model to derive both word-level and sentence-level semantics. Then the random walk with restart (RWR) is applied to extract subgraphs centered on each news, which are further fed to our proposed subgraph Transformer to quantify the authenticity. Extensive experiments on five real-world datasets demonstrate the superior performance of HeteroSGT over five baselines. Further case and ablation studies validate our motivation and demonstrate that performance improvement stems from our specially designed components.
Machine Unlearning (MU) algorithms have become increasingly critical due to the imperative adherence to data privacy regulations. The primary objective of MU is to erase the influence of specific data samples on a given model without the need to retrain it from scratch. Accordingly, existing methods focus on maximizing user privacy protection. However, there are different degrees of privacy regulations for each real-world web-based application. Exploring the full spectrum of trade-offs between privacy, model utility, and runtime efficiency is critical for practical unlearning scenarios. Furthermore, designing the MU algorithm with simple control of the aforementioned trade-off is desirable but challenging due to the inherent complex interaction. To address the challenges, we present Controllable Machine Unlearning (ConMU), a novel framework designed to facilitate the calibration of MU. The ConMU framework contains three integral modules: an important data selection module that reconciles the runtime efficiency and model generalization, a progressive Gaussian mechanism module that balances privacy and model generalization, and an unlearning proxy that controls the trade-offs between privacy and runtime efficiency. Comprehensive experiments on various benchmark datasets have demonstrated the robust adaptability of our control mechanism and its superiority over established unlearning methods. ConMU explores the full spectrum of the Privacy-Utility-Efficiency trade-off and allows practitioners to account for different real-world regulations. Source code available at: https://github.com/guangyaodou/ConMU
The results returned by image search engines have the power to shape peoples' perceptions about social groups. Existing work on image search engines leverages hand-selected queries for occupations like "doctor" and "engineer" to quantify racial and gender bias in search results. We complement this work by analyzing peoples' real-world image search queries and measuring the distributions of perceived gender, skin tone, and age in their results. We collect 54,070 unique image search queries and analyze 1,481 open-ended people queries (i.e. not queries for named entities) from a representative sample of 643 US residents. For each query, we analyze the top 15 results returned on both Google and Bing Images. Analysis of real-world image search queries produces multiple insights. First, less than 5% of unique queries are open-ended people queries. Second, fashion queries are, by far, the most common category of open-ended people queries, accounting for over 30% of the total. Third, the modal skin tone on the Monk Skin Tone scale is two out of ten (the second lightest) for images from both search engines. Finally, we observe a bias against older people: eleven of our top fifteen query categories have a median age that is lower than the median age in the US.
Can One Embedding Fit All? A Multi-Interest Learning Paradigm Towards Improving User Interest Diversity Fairness
PDF ↗Recommender systems (RSs) have gained widespread applications across various domains owing to the superior ability to capture users' interests. However, the complexity and nuanced nature of users' interests, which span a wide range of diversity, pose a significant challenge in delivering fair recommendations. In practice, user preferences vary significantly; some users show a clear preference toward certain item categories, while others have a broad interest in diverse ones. Even though it is expected that all users should receive high-quality recommendations, the effectiveness of RSs in catering to this disparate interest diversity remains under-explored. In this work, we investigate whether users with varied levels of interest diversity are treated fairly. Our empirical experiments reveal an inherent disparity: users with broader interests often receive lower-quality recommendations. To mitigate this, we propose a multi-interest framework that uses multiple (virtual) interest embeddings rather than single ones to represent users. Specifically, the framework consists of stacked multi-interest representation layers, which include an interest embedding generator that derives virtual interests from shared parameters, and a center embedding aggregator that facilitates multi-hop aggregation. Experiments demonstrate the effectiveness of the framework in achieving better trade-off between fairness and utility across various datasets and backbones.
This paper presents a study of GDPR compliance under the Interactive Advertising Bureau Europe's Transparency and Consent Framework (TCF). This framework provides digital advertising market participants a standard for sharing users' privacy consent choices. TCF is widely used across the Internet, and this paper presents a thorough experimental evaluation of both the compliance of websites with TCF and its impact on user privacy. We reviewed 2,230 websites that use TCF and accepted the automatic decline of user consent by our data collection system. Unlike previous work on GDPR compliance, we found that most websites using TCF properly record the user's consent choice. However, we found that 72.8% of the websites that were TCF compliant claimed legitimate interest as a rationale for overriding the consent choice. While legitimate interest is legal under GDPR, previous studies have shown that most users disagreed with how it is being used to collect data. Additionally, analysis of cookies set to the browsers indicates that TCF may not fully protect user privacy even when websites are compliant. Our research provides regulators and publishers with a data collection and analysis system to monitor compliance, detect non-compliance, and examine questionable practices of circumventing user consent choices using legitimate interest.
As a popular choice for video and entertainment streaming, YouTube hosts a large audience, including children, who form a growing proportion of its users. Despite separate "made for kids" labelling and stricter moderation of these videos, inappropriate advertising remains a concern as it threatens the safety of YouTube for young viewers. This paper is the first comparative measurement study that explores how advertisement exposure and content vary across child-oriented videos on YouTube. We do this by conducting a cross-regional advertisement analysis on highly viewed "made for kids" labelled content across a total of ten countries with varying regulation. A second front of comparison is carried out between ad patterns on unlabelled and labelled child-oriented videos. Our analysis reveals that the safety of a child's YouTube experience is shaped significantly by their external environment. There also appears to be lax enforcement of YouTube ad and child protection policies, indicated by the presence of unlabelled child-oriented content with weak ad regulation. We discuss the implications of inappropriate exposure on children and suggest policy and implementation measures to mitigate this threat.