A Study of Smoothing Methods for Language Models Applied to Ad Hoc Information Retrieval

By Chengxiang Zhai, John D. Lafferty. 2001

283 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems, Information Retrieval & Search, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Rodrigo Nogueira
2. Yiqun Liu
3. Shaoping Ma
4. Jiafeng Guo
5. Xueqi Cheng
6. Yixing Fan
7. W. Bruce Croft
8. Min Zhang

In Information Systems

1. Yiqun Liu
2. Shaoping Ma
3. Yongfeng Zhang
4. Min Zhang
5. Rodrigo Nogueira
6. Massimiliano Ciaramita
7. Jannis Bulian
8. Jiaxin Mao

In Information Retrieval & Search

1. Rodrigo Nogueira
2. Xueqi Cheng
3. Jiafeng Guo
4. Shaoping Ma
5. Yiqun Liu
6. Jiaxin Mao
7. Jingtao Zhan
8. Yixing Fan

In Natural Language Processing

1. Massimiliano Ciaramita
2. Jannis Bulian
3. Yixing Fan
4. Rodrigo Nogueira
5. Jiafeng Guo
6. Xueqi Cheng
7. Xinyu Ma
8. Jun Zhao

Citations in the corpus

• Representation Learning Using Multi-Task Deep Neural Networks for Semantic Classification and Information Retrieval (2015, neutral): “388 Unigram Language Model (Zhai and Lafferty, 2001) 0.”
• Multi-agent query reformulation: Challenges and the role of diversity (2019, methodology): “We use a Dirichlet smoothed language model (Zhai & Lafferty, 2001) to compute a language model from a document d ∈ D0: P (t|d) = tf(t, d) + uP (t|C) |d|+ u , (5) where u is a scalar constant (u = 1500 in our experiments), and P (t|C) is the probability of t occurring in the entire corpus C.”
• Leveraging Passage-level Cumulative Gain for Document Ranking (2020, background): “For many existing ranking methods, no matter whether they are unsupervised (e.g. BM25 [34] and language models [32, 42]) or supervised (e.g. learning to rank [1, 22] or deep ranking [27]), they usually capture relevance signals at the whole document level.”
• Learning Continuous Word Embedding with Metadata for Question Retrieval in Community Question Answering (2015, methodology): “Besides, we also introduce a baseline language model (LM) (Zhai and Lafferty, 2001) for comparison.”
• PROP: Pre-training with Representative Words Prediction for Ad-hoc Retrieval (2020, methodology): “onal retrieve models, pre-trained models, and previous stateof-the-art neural ranking models. For traditional retrieval models, we take two representative ranking methods: •QL: Query likelihood model [43] is one of the best performing language models based on Dirichlet smoothing. •BM25: The BM25 for”
• Asking Clarifying Questions in Open-Domain Information-Seeking Conversations (2019, unknown): “prior smoothing [52] where we linearly interpolate two likelihood models: one based on the original query, and one based on the questions and their respective answers.”
• Learning to Coordinate Multiple Reinforcement Learning Agents for Diverse Query Reformulation (2018, methodology): “We use a Dirichlet smoothed language model (Zhai & Lafferty, 2001) to compute a language model from a document d ∈ D0: P (t|d) = tf(t, d) + uP (t|C) |d|+ u , (5) where u is a scalar constant (u = 1500 in our experiments), and P (t|C) is the probability of t occurring in the entire corpus C.”
• Retrieval And Structuring Augmented Generation with Large Language Models (2025, unknown): “Other variants include BM25-adpt [155], LM-DS [284], and LM-PYP [175], each optimizing the BM25 formula for different retrieval scenarios.”