Advances in Information Retrieval: 44th European Conference on IR Research, ECIR 2022, Stavanger, Norway, April 10–14, 2022, Proceedings, Part I (Lecture Notes in Computer Science, 13185)

This two-volume set LNCS 13185 and 13186 constitutes the refereed proceedings of the 44th European Conference on IR Research, ECIR 2022, held in April 2022, due to the COVID-19 pandemic. The 35 full papers presented together with 11 reproducibility papers, 13 CLEF lab descriptions papers, 12 doctoral consortium papers, 5 workshop abstracts, and 4 tutorials abstracts were carefully reviewed and selected from 395 submissions. Preface Organization Contents – Part I Contents – Part II Full Papers Supercalifragilisticexpialidocious: Why Using the ``Right'' Readability Formula in Children's Web Search Matters 1 Introduction 2 Background and Related Work 3 The Fit of Readability Formulas on Web Text 4 The Effect of Readability on Web Search for Children 5 Conclusion and Future Work References PARM: A Paragraph Aggregation Retrieval Model for Dense Document-to-Document Retrieval 1 Introduction 2 Related Work 3 Paragraph Aggregation Retrieval Model (PARM) 3.1 Workflow 3.2 Vector-Based Aggregation with Reciprocal Rank Fusion Weighting (VRRF) 3.3 Training Strategies 4 Experiment Design 4.1 Training and Test Collections 4.2 Baselines 5 Results and Analysis 5.1 RQ1: VRRF Aggregation for PARM 5.2 RQ2: PARM VRRF vs Document-Level Retrieval 5.3 RQ3: Paragraph-Level vs Document-level Labelled Training 5.4 Analysis of Paragraph Relations 6 Conclusion References Recommendation of Compatible Outfits Conditioned on Style 1 Introduction 2 Methodology 2.1 Style Encoder Network 2.2 Style classifier 2.3 SCA Net 2.4 Outfit Generation 3 Experimental Evaluation 3.1 Compatibility Experiment 3.2 Style Experiments 4 Conclusion References Why Did You Not Compare with That? Identifying Papers for Use as Baselines 1 Introduction 2 Related Work 3 Dataset for Baseline Classification 3.1 Observations and Characteristics of the Dataset 4 Multi-module Attention Based Baseline Classifier 5 Empirical Results and Discussions 6 Conclusions References Exploring Entities in Event Detection as Question Answering 1 Introduction 2 Related Work 3 Event Question Answering Model with Entity Positions, Types, and Argument Roles 4 Experiments 5 Discussion 5.1 Trigger Ambiguity Analysis 5.2 Evaluation on Unseen Event Types 6 Conclusions and Perspectives References Validating Simulations of User Query Variants 1 Introduction 2 Related Work 3 Approach 3.1 Query Simulation 3.2 Evaluation Framework 3.3 Datasets and Implementation Details 4 Experimental Results 5 Discussion 6 Conclusion A Appendix References Out-of-Domain Semantics to the Rescue! Zero-Shot Hybrid Retrieval Models 1 Introduction 2 Related Work 3 Method 3.1 Hybrid Retrieval Model 3.2 Lexical Retrieval Model 3.3 Deep Retrieval Model 4 Experimental Setup 4.1 Datasets 4.2 Data Processing and Benchmarking 5 Evaluation 5.1 Generalization of the Deep Retrieval Model 5.2 Utility of Query and Document Expansion 5.3 Complementarity of Lexical and Deep Retrieval Models 5.4 Effectiveness of the Proposed Hybrid Model 6 Discussion 7 Conclusion References Incorporating Ranking Context for End-to-End BERT Re-ranking 1 Introduction 2 Related Work 3 Method 3.1 Overview 3.2 End-to-End Groupwise Scorer 3.3 Light-Weight Pseudo Relevance Feedback 3.4 End-to-End Training of the Model 4 Experiment Setup 4.1 Dataset and Metrics 4.2 Baselines and Co-BERT Variants 4.3 Model Training and Inference 5 Results 5.1 Overall Performance of Co-BERT 5.2 Study of Groupwise Ranking 5.3 Study of Light-Weight PRF Calibrator 5.4 Effectiveness on TREC DL 6 Conclusion References WANDS: Dataset for Product Search Relevance Assessment 1 Introduction 2 Related Work 3 Annotation Guidelines Design 3.1 Design Principles 3.2 Query-Product Annotation 4 Annotation Process 4.1 Query Sampling 4.2 Constructing the Product Pool 4.3 Iterative Product Mining for Dataset Completeness 5 Dataset 6 Experimental Evaluation 7 Discussion 7.1 Effectiveness of Sampling Sources 7.2 Iterative Product Mining for Dataset Completeness Step 8 Conclusions and Future Work References Searching, Learning, and Subtopic Ordering: A Simulation-Based Analysis 1 Introduction 2 Related Work 3 Subtopic-Aware Complex Searcher Model (SACSM) 4 Experimental Method 4.1 Fixed SACSM Components 4.2 Variable SACSM Components 4.3 Simulation Setup 5 Results 6 Conclusions References Immediate Text Search on Streams Using Apoptosic Indexes 1 Introduction 2 Background 3 Apoptosic Structures 3.1 Text Search 4 Experiments 5 Conclusions References Influence-Based Deep Network for Next POIs Prediction 1 Introduction 2 Problem Definition 3 Methodology 3.1 User Representation 3.2 Group Representation 3.3 The Interaction Learning Module 4 Experimental Evaluation 5 Conclusion References A Dependency-Aware Utterances Permutation Strategy to Improve Conversational Evaluation 1 Introduction 2 Related Work 3 Methodology 3.1 RQ1: A Dependence-Aware Utterance Permutation Strategy 3.2 RQ3: Exploiting Permuted Conversation Utterances 4 Experimental Analysis 4.1 Conversational Models 4.2 RQ1: Permuting Conversations 4.3 RQ2: Conversational Systems Performance on Permuted Conversations 4.4 RQ3: Comparing Systems via ANOVA 5 Conclusions and Future Works References Sentiment Guided Aspect Conditioned Dialogue Generation in a Multimodal System 1 Introduction 2 Related Work 3 Methodology 4 Dataset 5 Experiments 6 Results and Analysis 7 Conclusion and Future Work References An Analysis of Variations in the Effectiveness of Query Performance Prediction 1 Introduction 2 Related Work 3 Anatomy of a QPP Evaluation Framework 4 Experimental Setup 5 Results 5.1 RQ1: Variations in QPP Evaluations 5.2 RQ2: Variations in the Relative Ranks of QPP Methods 6 Concluding Remarks References Search Clarification Selection via Query-Intent-Clarification Graph Attention 1 Introduction 2 Related Work 2.1 Conversational Search 2.2 Asking Clarifications 3 Our Proposed Framework 3.1 Model Architecture 3.2 Loss Function 4 Experiments 4.1 Dataset 4.2 Baselines 4.3 Evaluation Metrics 4.4 Implementation Details 4.5 Experimental Results 5 Conclusion References Continual Learning of Long Topic Sequences in Neural Information Retrieval 1 Introduction 2 Related Works 3 Research Design for Continual Learning in IR 3.1 Continual Learning Setting and Metrics 3.2 Neural Ranking Models and Learning 4 MSMarco Continual Learning Corpus 4.1 RQ1: Modeling the Long Topic Sequence 4.2 Evaluating the Long Topic Sequence 4.3 IR-Driven Controlled Stream-Based Scenario 5 Model Performance and Learning Behavior on Long Topic Sequences 5.1 RQ2: Performances on the MSMarco Long Topic Sequence 5.2 Fine-Grained Analysis 5.3 RQ3: Behavior on IR-Driven Controlled Settings 6 Conclusion and Future Work References Ensemble Model Compression for Fast and Energy-Efficient Ranking on FPGAs 1 Introduction 2 Related Work 3 Using Programmable Logic for Ranking 4 Ranking Model Compression 5 Experiments 5.1 Effectiveness Assessment 5.2 Efficiency Assessment 5.3 Data Transfer Assessment 5.4 Energy Consumption Assessment 6 Conclusions and Future Work References Local Citation Recommendation with Hierarchical-Attention Text Encoder and SciBERT-Based Reranking 1 Introduction 2 Related Work 3 Proposed Dataset 4 Approach 4.1 Prefetching Model 4.2 Reranking Model 4.3 Loss Function 5 Experiments 6 Results and Discussion 6.1 Prefetching Results 6.2 Reranking Results 6.3 Performance of Entire Recommendation Pipeline 7 Conclusion References Extending CLIP for Category-to-Image Retrieval in E-Commerce 1 Introduction 2 Related Work 3 Approach 4 Experimental Setup 5 Experimental Results 6 Error Analysis 7 Conclusion References WIDAR - Weighted Input Document Augmented ROUGE 1 Introduction 2 Related Works 3 WIDAR Evaluation Metric 3.1 Weighted ROUGE 3.2 Combining Weighted ROUGE with Input Document Similarity 4 Experiments 4.1 Dataset 4.2 Evaluation of Evaluation Metric 4.3 Experimental Settings 5 Results and Discussions 5.1 Computational Time Analysis 5.2 Ablation Study 5.3 Study of Human Judgement Scores 6 Conclusion References Bi-granularity Adversarial Training for Non-factoid Answer Retrieval 1 Introduction 2 Related Work 3 Methodology 3.1 Task Definition 3.2 Basic Answer Retrieval Model (Baseline) 3.3 Adversarial Examples Generation Strategy 3.4 Self-adaptive Adversarial Training 4 Experimentation 4.1 Data and Evaluation 4.2 Hyperparameter Settings and Ranking Scenario 5 Results and Discussion 5.1 Utility of Token-Level Adversarial Examples 5.2 Effect of Bi-granularity Adversarial Training (BAT) 5.3 Comparison to State of the Art 5.4 Statistical and Practical Significance Test 6 Conclusion References RATE: A Reliability-Aware Tester-Based Evaluation Framework of User Simulators 1 Introduction 2 Tester-Based Evaluation (TBE) 3 Reliability-Aware Tester-Based Evaluation (RATE) 3.1 Reliability of a Tester 3.2 Reliability of User Simulator 4 Evaluation 4.1 Experiment Design 4.2 Experiment Results 5 Conclusion References HC4: A New Suite of Test Collections for Ad Hoc CLIR 1 Introduction 2 Related Work 3 Collection Development Methodology 3.1 Topic Development 3.2 Relevance Judgments 3.3 Contemporaneous Reports 4 Collection Details 4.1 Development and Annotation Time 4.2 Inter-assessor Agreement 5 Baseline Runs 6 Conclusion References Did I See It Before? Detecting Previously-Checked Claims over Twitter 1 Introduction 2 Related Work 3 The Proposed Pipeline 3.1 Preprocessing 3.2 Initial Retrieval 3.3 Reranking 4 Experimental Setup 5 Experimental Evaluation 5.1 Initial Retrieval with Preprocessing ([rqspspreprocessing]RQ1) 5.2 MonoBERT for Reranking ([rqspstunespsdepth]RQ2) 5.3 Leveraging Verified Claim Fields ([rqspsfields]RQ3) 5.4 Performance on Arabic Data ([rqspsarabic]RQ4) 6 Conclusion and Future Work References Transfer Learning Approaches for Building Cross-Language Dense Retrieval Models 1 Introduction 2 Related Work 3 ColBERT-X 3.1 CLIR Training Strategies 3.2 Retrieval 4 Experiments 5 Detailed Analysis 5.1 Effect of Machine Translation 5.2 Effect of Multilingual Language Models 5.3 Pseudo-Relevance Feedback 5.4 Effect of Longer Queries 5.5 Indexing Footprint 6 Conclusion References Evaluating the Robustness of Retrieval Pipelines with Query Variation Generators 1 Introduction 2 Related Work 3 Automatic Query Variations 3.1 UQV Taxonomy 3.2 Query Generators 4 Experimental Setup 5 Results 6 Conclusions References Exploiting Document-Based Features for Clarification in Conversational Search 1 Introduction 2 Related Work 3 Methodology 3.1 Feature Extraction from Retrieved Documents 3.2 Facet Extraction 3.3 Facet-Based Question Generation 4 Evaluation Setting 4.1 Evaluation Dataset 4.2 Evaluation of Document-Level Features 4.3 Evaluation of Facet Extraction 4.4 Evaluation of the Usefulness of Facet-Based Clarifying Questions 5 Results and Discussion 5.1 Precision and Recall of Extracted Feature Sets 5.2 Facet-Based Retrieval 5.3 Usefulness of Clarifying Questions 6 Conclusions References Adversarial Multi-task Model for Emotion, Sentiment, and Sarcasm Aided Complaint Detection 1 Introduction 2 Related Work 3 Dataset 3.1 Data Annotation 3.2 Annotation Specifications 4 Proposed Methodology 4.1 Problem Definition 4.2 Multi-task Model for Emotion, Sentiment, and Sarcasm Aided Complaint Detection (MTLAll) 4.3 Feature Extraction 4.4 Encoding Model 4.5 Intra-model Attention 4.6 Adversarial Loss 4.7 Output Layer 5 Experiments, Results, and Analysis 5.1 Baselines 5.2 Experimental Setup 5.3 Results and Discussion 5.4 Error Analysis 6 Conclusion and Future Work References Joint Personalized Search and Recommendation with Hypergraph Convolutional Networks 1 Introduction 2 Related Work 3 Problem Definition 4 Method 4.1 Hypergraph Construction 4.2 HyperSaR Convolution Operation 4.3 Model Training and Inference 5 Experimental Setup 5.1 Datasets 5.2 Baselines 5.3 Hyperparameter Setting 6 Experiment Results 6.1 Evaluation on the JPSR Task 6.2 Hyperparameter Impact 7 Conclusion References Topic Aware Contextualized Embeddings for High Quality Phrase Extraction 1 Introduction 2 Related Work 2.1 Unsupervised Keyphrase Extraction 2.2 Contextualized Vector Representations 3 Methodology 3.1 Vector Representations for the Phrases and the Document 3.2 Graph Based Ranking 4 Experiments and Results 4.1 Datasets 4.2 Baselines and Variants of the Proposed Method 4.3 Results and Discussion 4.4 Effects of Different Hyperparameters 4.5 Keyphrase Expansion Results 5 Conclusions References Topic Modeling on Podcast Short-Text Metadata 1 Introduction 2 Related Work 3 Methods 3.1 Notations and Preliminaries 3.2 NE-informed Corpus Embedding (NEiCE) 4 Datasets 5 Experimental Setup 6 Results and Discussion 7 Conclusion References Effective Rating Prediction Using an Attention-Based User Review Sentiment Model 1 Introduction 2 Related Work 3 The SentiAttn Model 3.1 Task Definition 3.2 Review Sentiment Information Analysis 3.3 Model Architecture 4 Experimental Setup 4.1 Datasets 4.2 Baselines and Evaluation Metrics 4.3 Model Setting 5 Results 6 Conclusions References Goldilocks: Just-Right Tuning of BERT for Technology-Assisted Review 1 Introduction 2 Background 3 Adapting BERT for TAR 4 Experiment Setup 4.1 Data Sets 4.2 Software and Evaluation 4.3 Hardware 5 Results and Analysis 5.1 Language Model Fine-Tuning 5.2 Just-Right Varies Across Tasks 5.3 Running Time 6 Summary and Future Works References Multi-modal Sentiment and Emotion Joint Analysis with a Deep Attentive Multi-task Learning Model 1 Introduction 2 Related Work 2.1 Multi-modal Sentiment Analysis 2.2 Multi-modal Emotion Recognition 3 Methodology 3.1 Task Description and Overall Network 3.2 Multi-modal Encoder 3.3 External Information Extraction 3.4 Multi-head Cross-modal Attentive Fusion 3.5 Cross-task Attention Mechanism 3.6 Classification 4 Experimental Setup 4.1 Experiment Settings 4.2 Comparison Models 4.3 Comparative Analysis 4.4 STL v/s MTL Framework 4.5 Ablation Test 4.6 Misclassification Cases 5 Conclusions and Future Work References Reproducibility Papers Do Lessons from Metric Learning Generalize to Image-Caption Retrieval? 1 Introduction 2 Background and Related Work 3 Do Findings from Metric Learning Extend to ICR? 4 A Method for Analyzing the Behavior of Loss Functions 5 Analyzing the Behavior of Loss Functions for ICR 6 Conclusion References Consumer Fairness in Recommender Systems: Contextualizing Definitions and Mitigations 1 Introduction 2 Research Methodology 2.1 Mitigation Procedures Collection 2.2 Mitigation Procedures Reproduction 2.3 Mitigation Procedures Evaluation 3 Experimental Results 3.1 Impact on Recommendation Utility (RQ1) 3.2 Impact on Group Unfairness (RQ2) 3.3 Relationships Between Representation and Unfairness (RQ3) 4 Discussion and Conclusions References The Power of Anchor Text in the Neural Retrieval Era 1 Introduction 2 Related Work 3 The Webis MS MARCO Anchor Text 2022 Dataset 4 Properties of Anchor Texts, Queries, and Documents 5 Anchor Text and Retrieval Effectiveness 5.1 Navigational Queries for MS MARCO 5.2 Retrieval Models and Training 5.3 Evaluation 6 Conclusion References Automation of Citation Screening for Systematic Literature Reviews Using Neural Networks: A Replicability Study 1 Introduction 2 Related Work 3 Experiment Setup 3.1 Models 3.2 Data 3.3 Evaluation 3.4 Code 4 Results and Discussion 4.1 Replicability Study 4.2 Impact of Input Features 4.3 Training Time 4.4 Precision@95%recall 5 Conclusions References Improving Query Representations for Dense Retrieval with Pseudo Relevance Feedback: A Reproducibility Study 1 Introduction 2 Related Work 3 Improving Query Representations for Dense Retrievers with Pseudo Relevance Feedback 4 Experimental Settings 4.1 Datasets 4.2 Models 4.3 Inference and Training 4.4 Evaluation Metrics 4.5 Research Questions 5 Results and Analysis 5.1 RQ1: Reproduce ANCE-PRF Inference 5.2 RQ2: Replicate ANCE-PRF Training 5.3 RQ3: Generalisability of ANCE-PRF Beyond ANCE 6 Conclusion References Another Look at DPR: Reproduction of Training and Replication of Retrieval 1 Introduction 2 Methods 2.1 Retriever 2.2 Reader 2.3 Final Evidence Fusion 3 Experimental Setup 4 Results 4.1 Reproduction of Training 4.2 Replication of Retrieval 4.3 Replication of End-to-End QA 5 Conclusion References Reproducing Personalised Session Search Over the AOL Query Log 1 Introduction 2 Background 3 Reconstructing the AOL Document Corpus 4 Comparing AOLIA with AOL17 5 Reproduction and Replication 5.1 Methods 5.2 Experimental Settings 5.3 Results 6 Limitations 7 Conclusions References Revisiting Popularity and Demographic Biases in Recommender Evaluation and Effectiveness 1 Introduction 2 Related Work 3 Demographics and Popularity 4 Method 4.1 Datasets 4.2 Model 4.3 Experiment Protocol 5 Results 5.1 Impact on Age 5.2 Impact on Gender 5.3 Impact on Country 6 Discussion and Conclusion References Squeezing Water from a Stone: A Bag of Tricks for Further Improving Cross-Encoder Effectiveness for Reranking 1 Introduction 2 Related Work 2.1 Cross-Encoders 2.2 Bi-Encoders 2.3 Hard Negatives 2.4 Pretrained Transformers for Cross-Encoders 3 Loss Functions 3.1 Cross Entropy and Hinge Loss 3.2 Localized Contrastive Estimation 4 Experimental Setup 4.1 Data 4.2 Training and Inference 5 Results and Discussion 5.1 Loss Functions 5.2 In-distributional Training Example and Hard Negative 5.3 LCE Group Size 6 Conclusion References An Evaluation Study of Generative Adversarial Networks for Collaborative Filtering 1 Introduction 2 Collaborative Filtering Generative Adversarial Networks 3 CFGAN Theoretical and Methodological Questions 3.1 Real Profiles as Condition Vectors 3.2 No Random Noise 3.3 Methodological Questions 4 Experimental Methodology 4.1 Methodology for the Replicability of CFGAN 4.2 Methodology for the Reproducibility of CFGAN 5 Experiments Results and Discussion 5.1 RQ1: CFGAN Replicability and Numerical Stability 5.2 RQ2: Impact of Theoretical and Methodological Concerns 5.3 RQ3: Reproducibility Evaluation Against Properly Tuned Baselines 6 Conclusions References Seed-Driven Document Ranking for Systematic Reviews: A Reproducibility Study 1 Introduction 2 Replicating SDR 2.1 Document Representation 2.2 Term Weighting 2.3 Document Scoring 2.4 Multi-SDR 3 Experimental Setup 3.1 Datasets 3.2 Baselines 3.3 Evaluation Measures 3.4 Document Pre-processing 4 Results 4.1 Generalisability of SDR 4.2 Effect of Multiple Seed Studies 4.3 Variability of Seed Studies on Effectiveness 5 Related Work 6 Conclusions References Author Index