EALM: Introducing Multidimensional Ethical Alignment in Conversational Information Retrieval

Artificial intelligence (AI) technologies should adhere to human norms to better serve our society and avoid disseminating harmful or misleading information, particularly in Conversational Information Retrieval (CIR). Previous work, including approaches and datasets, has not always been successful or sufficiently robust in taking human norms into consideration. To this end, we introduce a workflow that integrates ethical alignment, with an initial ethical judgment stage for efficient data screening. To address the need for ethical judgment in CIR, we present the QA-ETHICS dataset, adapted from the ETHICS benchmark, which serves as an evaluation tool by unifying scenarios and label meanings. However, each scenario only considers one ethical concept. Therefore, we introduce the MP-ETHICS dataset to evaluate a scenario under multiple ethical concepts, such as justice and Deontology. In addition, we suggest a new approach that achieves top performance in both binary and multi-label ethical judgment tasks. Our research provides a practical method for introducing ethical alignment into the CIR workflow. The data and code are available at https://github.com/wanng-ide/ealm .

Open-Domain Dialogue Quality Evaluation: Deriving Nugget-level Scores from Turn-level Scores

Existing dialogue quality evaluation systems can return a score for a given system turn from a particular viewpoint, e.g., engagingness. However, to improve dialogue systems by locating exactly where in a system turn potential problems lie, a more fine-grained evaluation may be necessary. We therefore propose an evaluation approach where a turn is decomposed into nuggets (i.e., expressions associated with a dialogue act), and nugget-level evaluation is enabled by leveraging an existing turn-level evaluation system. We demonstrate the potential effectiveness of our evaluation method through a case study.

Towards Consistency Filtering-Free Unsupervised Learning for Dense Retrieval

Domain transfer is a prevalent challenge in modern neural Information Retrieval (IR). To overcome this problem, previous research has utilized domain-specific manual annotations and synthetic data produced by consistency filtering to finetune a general ranker and produce a domain-specific ranker. However, training such consistency filters are computationally expensive, which significantly reduces the model efficiency. In addition, consistency filtering often struggles to identify retrieval intentions and recognize query and corpus distributions in a target domain. In this study, we evaluate a more efficient solution: replacing the consistency filter with either direct pseudo-labeling, pseudo-relevance feedback, or unsupervised keyword generation methods for achieving consistent filtering-free unsupervised dense retrieval. Our extensive experimental evaluations demonstrate that, on average, TextRank-based pseudo relevance feedback outperforms other methods. Furthermore, we analyzed the training and inference efficiency of the proposed paradigm. The results indicate that filtering-free unsupervised learning can continuously improve training and inference efficiency while maintaining retrieval performance. In some cases, it can even improve performance based on particular datasets.

A Meta-Evaluation of C/W/L/A Metrics: System Ranking Similarity, System Ranking Consistency and Discriminative Power

Recently, Moffat et al. proposed an analytic framework, namely C/W/L/A, for offline evaluation metrics. This framework allows information retrieval (IR) researchers to design evaluation metrics through the flexible combination of user browsing models and user gain aggregations. However, the statistical stability of C/W/L/A metrics with different aggregations is not yet investigated. In this study, we investigate the statistical stability of C/W/L/A metrics from the perspective of: (1) the system ranking similarity among aggregations, (2) the system ranking consistency of aggregations and (3) the discriminative power of aggregations. More specifically, we combined various aggregation functions with the browsing model of Precision, Discounted Cumulative Gain (DCG), Rank-Biased Precision (RBP), INST, Average Precision (AP) and Expected Reciprocal Rank (ERR), examing their performances in terms of system ranking similarity, system ranking consistency and discriminative power on two offline test collections. Our experimental result suggests that, in terms of system ranking consistency and discriminative power, the aggregation function of expected rate of gain (ERG) has an outstanding performance while the aggregation function of maximum relevance usually has an insufficient performance. The result also suggests that Precision, DCG, RBP, INST and AP with their canonical aggregation all have favourable performances in system ranking consistency and discriminative power; but for ERR, replacing its canonical aggregation with ERG can further strengthen the discriminative power while obtaining a system ranking list similar to the canonical version at the same time.

SWAN: A Generic Framework for Auditing Textual Conversational Systems

We present a simple and generic framework for auditing a given textual conversational system, given some samples of its conversation sessions as its input. The framework computes a SWAN (Schematised Weighted Average Nugget) score based on nugget sequences extracted from the conversation sessions. Following the approaches of S-measure and U-measure, SWAN utilises nugget positions within the conversations to weight the nuggets based on a user model. We also present a schema of twenty (+1) criteria that may be worth incorporating in the SWAN framework. In our future work, we plan to devise conversation sampling methods that are suitable for the various criteria, construct seed user turns for comparing multiple systems, and validate specific instances of SWAN for the purpose of preventing negative impacts of conversational systems on users and society. This paper was written while preparing for the ICTIR 2023 keynote (to be given on July 23, 2023).

A First Look at LLM-Powered Generative News Recommendation

Personalized news recommendation systems have become essential tools for users to navigate the vast amount of online news content, yet existing news recommenders face significant challenges such as the cold-start problem, user profile modeling, and news content understanding. Previous works have typically followed an inflexible routine to address a particular challenge through model design, but are limited in their ability to understand news content and capture user interests. In this paper, we introduce GENRE, an LLM-powered generative news recommendation framework, which leverages pretrained semantic knowledge from large language models to enrich news data. Our aim is to provide a flexible and unified solution for news recommendation by moving from model design to prompt design. We showcase the use of GENRE for personalized news generation, user profiling, and news summarization. Extensive experiments with various popular recommendation models demonstrate the effectiveness of GENRE. We will publish our code and data for other researchers to reproduce our work.

NER-to-MRC: Named-Entity Recognition Completely Solving as Machine Reading Comprehension

Named-entity recognition (NER) detects texts with predefined semantic labels and is an essential building block for natural language processing (NLP). Notably, recent NER research focuses on utilizing massive extra data, including pre-training corpora and incorporating search engines. However, these methods suffer from high costs associated with data collection and pre-training, and additional training process of the retrieved data from search engines. To address the above challenges, we completely frame NER as a machine reading comprehension (MRC) problem, called NER-to-MRC, by leveraging MRC with its ability to exploit existing data efficiently. Several prior works have been dedicated to employing MRC-based solutions for tackling the NER problem, several challenges persist: i) the reliance on manually designed prompts; ii) the limited MRC approaches to data reconstruction, which fails to achieve performance on par with methods utilizing extensive additional data. Thus, our NER-to-MRC conversion consists of two components: i) transform the NER task into a form suitable for the model to solve with MRC in a efficient manner; ii) apply the MRC reasoning strategy to the model. We experiment on 6 benchmark datasets from three domains and achieve state-of-the-art performance without external data, up to 11.24% improvement on the WNUT-16 dataset.

Relevance Assessments for Web Search Evaluation: Should We Randomise or Prioritise the Pooled Documents? (CORRECTED VERSION)

In the context of depth-$k$ pooling for constructing web search test collections, we compare two approaches to ordering pooled documents for relevance assessors: the prioritisation strategy (PRI) used widely at NTCIR, and the simple randomisation strategy (RND). In order to address research questions regarding PRI and RND, we have constructed and released the WWW3E8 data set, which contains eight independent relevance labels for 32,375 topic-document pairs, i.e., a total of 259,000 labels. Four of the eight relevance labels were obtained from PRI-based pools; the other four were obtained from RND-based pools. Using WWW3E8, we compare PRI and RND in terms of inter-assessor agreement, system ranking agreement, and robustness to new systems that did not contribute to the pools. We also utilise an assessor activity log we obtained as a byproduct of WWW3E8 to compare the two strategies in terms of assessment efficiency.

Corrected Evaluation Results of the NTCIR WWW-2, WWW-3, and WWW-4 English Subtasks

Unfortunately, the official English (sub)task results reported in the NTCIR-14 WWW-2, NTCIR-15 WWW-3, and NTCIR-16 WWW-4 overview papers are incorrect due to noise in the official qrels files; this paper reports results based on the corrected qrels files. The noise is due to a fatal bug in the backend of our relevance assessment interface. More specifically, at WWW-2, WWW-3, and WWW-4, two versions of pool files were created for each English topic: a PRI ("prioritised") file, which uses the NTCIRPOOL script to prioritise likely relevant documents, and a RND ("randomised") file, which randomises the pooled documents. This was done for the purpose of studying the effect of document ordering for relevance assessors. However, the programmer who wrote the interface backend assumed that a combination of a topic ID and a document rank in the pool file uniquely determines a document ID; this is obviously incorrect as we have two versions of pool files. The outcome is that all the PRI-based relevance labels for the WWW-2 test collection are incorrect (while all the RND-based relevance labels are correct), and all the RND-based relevance labels for the WWW-3 and WWW-4 test collections are incorrect (while all the PRI-based relevance labels are correct). This bug was finally discovered at the NTCIR-16 WWW-4 task when the first seven authors of this paper served as Gold assessors (i.e., topic creators who define what is relevant) and closely examined the disagreements with Bronze assessors (i.e., non-topic-creators; non-experts). We would like to apologise to the WWW participants and the NTCIR chairs for the inconvenience and confusion caused due to this bug.

Zero-Shot Learners for Natural Language Understanding via a Unified Multiple Choice Perspective

We propose a new paradigm for zero-shot learners that is format agnostic, i.e., it is compatible with any format and applicable to a list of language tasks, such as text classification, commonsense reasoning, coreference resolution, and sentiment analysis. Zero-shot learning aims to train a model on a given task such that it can address new learning tasks without any additional training. Our approach converts zero-shot learning into multiple-choice tasks, avoiding problems in commonly used large-scale generative models such as FLAN. It not only adds generalization ability to models but also significantly reduces the number of parameters. Our method shares the merits of efficient training and deployment. Our approach shows state-of-the-art performance on several benchmarks and produces satisfactory results on tasks such as natural language inference and text classification. Our model achieves this success with only 235M parameters, which is substantially smaller than state-of-the-art models with billions of parameters. The code and pre-trained models are available at https://github.com/IDEA-CCNL/Fengshenbang-LM .