Leveraging Similar Users for Personalized Language Modeling with Limited Data

Welch, Charles; Gu, C.; Kummerfeld, Jonathan; Pérez-Rosas, Verónica; Mihalcea, Rada

doi:10.18653/v1/2022.acl-long.122

Cited by 10 publications

(7 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In contrast to Welch et al (2022a), we found that authorship attribution representations can scale to a large number of users by learning a projection layer to reduce it to a similar size as the text encoding. Contrary to King and Cook (2020), who found that priming outperformed other methods in relatively low data settings (like ours), we find that it underperforms other methods at the verdict-level.…”

Section: Resultsmentioning

confidence: 85%

“…Welch et al (2020b) modeled users by learning separate embedding matrices for each user in a shared embedding space. Welch et al (2022a) explored how to model users based on their similarity to others. They used the perplexity of personalized models and the predictions of an authorship attribution classifier to generate user representations.…”

Section: Personalizationmentioning

confidence: 99%

See 1 more Smart Citation

Unifying Data Perspectivism and Personalization: An Application to Social Norms

Plepi¹,

Neuendorf²,

Flek³

et al. 2022

Preprint

View full text Add to dashboard Cite

Instead of using a single ground truth for language processing tasks, several recent studies have examined how to represent and predict the labels of the set of annotators. However, often little or no information about annotators is known, or the set of annotators is small. In this work, we examine a corpus of social media posts about conflict from a set of 13k annotators and 210k judgements of social norms. We provide a novel experimental setup that applies personalization methods to the modeling of annotators and compare their effectiveness for predicting the perception of social norms. We further provide an analysis of performance across subsets of social situations that vary by the closeness of the relationship between parties in conflict, and assess where personalization helps the most.

show abstract

Section: Resultsmentioning

confidence: 85%

Section: Personalizationmentioning

confidence: 99%

Unifying Data Perspectivism and Personalization: An Application to Social Norms

Plepi¹,

Neuendorf²,

Flek³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…We discuss these decisions in §5.1. Nonetheless, in the case of scarce data on a single user's preferences, personalised LLMs may leverage similar users [233] or make inferences about their preferences and values from limited information. Making assumptions about the user (especially if they are demographically or geographically-informed) is a form of algorithmic profiling, risking the non-consensual categorisation of peoples [220].…”

Section: Ir4 Essentialism and Profilingmentioning

confidence: 99%

Personalisation within bounds: A risk taxonomy and policy framework for the alignment of large language models with personalised feedback

Kirk¹,

Vidgen²,

Röttger³

et al. 2023

Preprint

View full text Add to dashboard Cite

Large language models (LLMs) are used to generate content for an increasingly wide range of tasks, and are set to reach a growing audience in coming years due to integration in product interfaces like ChatGPT or search engines like Bing. This intensifies the need to ensure that models are aligned with human preferences and do not produce unsafe, inaccurate or toxic outputs. While alignment techniques like reinforcement learning with human feedback (RLHF) and red-teaming can mitigate some safety concerns and improve model capabilities, it is unlikely that an aggregate fine-tuning process can adequately represent the full range of users' preferences and values. Different people may legitimately disagree on their preferences for language and conversational norms, as well as on values or ideologies which guide their communication. Personalising LLMs through micro-level preference learning processes may result in models that are better aligned with each user. However, there are several normative challenges in defining the bounds of a societally-acceptable and safe degree of personalisation. In this paper, we ask how, and in what ways, LLMs should be personalised. First, we review literature on current paradigms for aligning LLMs with human feedback, and identify issues including (i) a lack of clarity regarding what alignment means; (ii) a tendency of technology providers to prescribe definitions of inherently subjective preferences and values; and (iii) a "tyranny of the crowdworker", exacerbated by a lack of documentation in who we are really aligning to. Second, we present a taxonomy of benefits and risks associated with personalised LLMs, for individuals and society at large. Finally, we propose a three-tiered policy framework that allows users to experience the benefits of personalised alignment, while restraining unsafe and undesirable LLM-behaviours within (supra-)national and organisational bounds.

show abstract

“…One might think, for example, that restricting a named entity to being comparable only to another named entity cannot damage the results, but experience has shown that this discrimination leads to a bad similarity between expressions such as "the Japanese president… 'And' in Japan, the president…" [26]. The system whose results are given in the evaluations thus operates without any restriction as to the syntactic nature of the terms compared [27]. Below is an example of calculating the similarity between two sentences.…”

Section: Similarity Of Mihalcia [23]mentioning

confidence: 99%

Realization of an intelligent evaluation system

Maarouf

Ayachi

Bahaj

2023

IJ-ICT

View full text Add to dashboard Cite

<span lang="EN-US">A number of benefits have been reported for computer-based assessments over traditional paper-based exams, both in terms of IT support for question development, reduced distribution and test administration costs, and automated support. Possible for the ranking. However, existing computerized assessment systems do not provide all kinds of questions, namely open questions that require writing solutions. To overcome the challenges of the existing, the objective of this work is to achieve an intelligent evaluation system (IES) responding to the problems identified, and which adapts to the different types of questions, especially open-ended questions of which the answer requires sentence writing or programming.</span>

show abstract

Leveraging Similar Users for Personalized Language Modeling with Limited Data

Cited by 10 publications

References 0 publications

Unifying Data Perspectivism and Personalization: An Application to Social Norms

Unifying Data Perspectivism and Personalization: An Application to Social Norms

Personalisation within bounds: A risk taxonomy and policy framework for the alignment of large language models with personalised feedback

Realization of an intelligent evaluation system

Contact Info

Product

Resources

About