Neural Ranking Models with Weak Supervision

By Mostafa Dehghani, Hamed Zamani, Aliaksei Severyn, Jaap Kamps, W. Bruce Croft. 2017

437 citations recorded by the source corpus.

Despite the impressive improvements achieved by unsupervised deep neural networks in computer vision and NLP tasks, such improvements have not yet been observed in ranking for information retrieval. The reason may be the complexity of the ranking problem, as it is not obvious how to learn from queries and documents when no supervised signal is available. Hence, in this paper, we propose to train a neural ranking model using weak supervision, where labels are obtained automatically without human annotators or any external resources (e.g., click data). To this aim, we use the output of an unsupervised ranking model, such as BM25, as a weak supervision signal. We further train a set of simple yet effective ranking models based on feed-forward neural networks. We study their effectiveness under various learning scenarios (point-wise and pair-wise models) and using different input representations (i.e., from encoding query-document pairs into dense/sparse vectors to using word embedding representation). We train our networks using tens of millions of training instances and evaluate it on two standard collections: a homogeneous news collection (Robust) and a heterogeneous large-scale web collection (ClueWeb). Our experiments indicate that employing proper objective functions and letting the networks to learn the input representation based on weakly supervised data leads to impressive performance, with over 13% and 35% MAP improvements over the BM25 model on the Robust and the ClueWeb collections. Our findings also suggest that supervised neural ranking models can greatly benefit from pre-training on large amounts of weakly labeled data that can be easily obtained from unsupervised IR models.

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

In Information Systems

In Information Retrieval & Search

In Natural Language Processing

Citations in the corpus

Distilling Knowledge from Reader to Retriever for Question Answering (2020, methodology): “We consider two other objectives which have been used in Dehghani et al. (2017), where BM25 is used as a teacher model to train a neural ranker.”
Revealing the Importance of Semantic Retrieval for Machine Reading at Scale (2019, background): “Information Retrieval Success in deep neural networks inspires their application to information retrieval (IR) tasks (Huang et al., 2013; Guo et al., 2016; Mitra et al., 2017; Dehghani et al., 2017).”
Relevance-based Word Embedding (2017, background): “More recently, Dehghani et al. [9] trained neural ranking models with weak supervision data (i.e., a set of noisy training data automatically generated by an existing unsupervised model) that learn word representations in an end-to-end ranking scenario.”
Context-Aware Sentence/Passage Term Importance Estimation For First Stage Retrieval (2019, background): “As discussed above, in both document and query term weighting problems, word embeddings have gained increasing attention due to their ability to encode certain aspects of text meaning [10, 13, 36].”
Transformer Memory as a Differentiable Search Index (2022, background): “Dual encoders (Dehghani et al., 2017; Gillick et al., 2018; Gao et al., 2021; Ni et al., 2021; Karpukhin et al., 2020) are a well-established paradigm for retrieval.”
Context-Aware Document Term Weighting for Ad-Hoc Search (2020, methodology): “Most of prior work uses dense text representations that enable a query to match every document to some degree, which makes them impractical for first-stage ranking in large-scale datasets [11, 13, 16, 18, 39].”
Efficient Document Re-Ranking for Transformers by Precomputing Term Representations (2020, methodology): “We train all transformer models using pairwise softmax loss [7] and the Adam optimizer [23] with a learning rate of 2×10−5.”
Multi-step Entity-centric Information Retrieval for Multi-Hop Question Answering (2019, neutral): “Neural ranking models have shown great potential and have been widely adopted in the IR community (Dehghani et al., 2017; Guo et al., 2019; Mitra et al., 2017; Zamani et al., 2018, inter-alia).”
Zero-shot Neural Retrieval via Domain-targeted Synthetic Query Generation (2020, background): “Previous work attempts to alleviate this issue by pre-training deep retrieval models on weakly supervised data such as user clicks, search results from traditional IR systems, or anchor texts (Borisov et al., 2016; Dehghani et al., 2017).”
To Clean or Not to Clean (2018, methodology): “The filtered version was created by eliminating all documents with a spam score of less than 70% (as suggested in https://plg.uwaterloo.ca/ ∼ gvcormac/clueweb09spam/ and used in References [28, 54, 71, 78]).”
Interactive Weak Supervision: Learning Useful Heuristics for Data Labeling (2020, background): “Further work in the growing body of literature on learning from multiple weak supervision sources has studied weak supervision for neural networks in information retrieval (Dehghani et al., 2017; Zamani et al., 2018; Zamani & Croft, 2018), the modeling of dependencies among heuristics in data progra”
Word Embedding Causes Topic Shifting; Exploit Global Context! (2017, methodology): “Word embedding is used in various IR tasks e.g. document retrieval [11, 12, 18], neural network-based retrieval models [4, 6, 8], and query expansion [16].”
Fidelity-Weighted Learning (2017, methodology): “For the ranking task, the employed student is proposed in (Dehghani et al., 2017b).”
The Role of Massively Multi-Task and Weak Supervision in Software 2.0 (2019, background): “…weak supervision [28] to create their training data for tasks involving text, image, video, time series, and other data, including at major organizations with substantial resources such as Google [10, 11, 20], Facebook [21], and Microsoft [16].”
PROP: Pre-training with Representative Words Prediction for Ad-hoc Retrieval (2020, background): “There might be some confusion between the proposed pre-training model PROP and those weak supervision methods [2, 11, 22] in”