More in related work

report/main.bib

@@ -122,3 +122,19 @@ @article{chen2022pseudo
   year={2022},
   publisher={Hindawi}
 }
+
+@article{xiong2020ance,
+  title={Approximate nearest neighbor negative contrastive learning for dense text retrieval},
+  author={Xiong, Lee and Xiong, Chenyan and Li, Ye and Tang, Kwok-Fung and Liu, Jialin and Bennett, Paul and Ahmed, Junaid and Overwijk, Arnold},
+  journal={arXiv preprint arXiv:2007.00808},
+  year={2020}
+}
+
+@inproceedings{yan2021waterlooclarke,
+  title={WaterlooClarke at the TREC 2021 conversational assistant track},
+  author={Yan, Xinyi and Clarke, CL and Arabzadeh, Negar},
+  booktitle={The Thirtieth Text REtrieval Conference Proceedings, TREC},
+  volume={21},
+  year={2021}
+}
+

report/main.tex

@@ -92,11 +92,14 @@ \subsection*{\texttt{doc2query} Document Expansion}\label{sec:doc2query}
 A notable solution to this is the \texttt{doc\-2query} method proposed by Nogueira et al. \cite{nogueira2019document}. It employs a text-to-text transformer to convert documents into queries. By generating and appending a few of these transformed queries to the original document, classical retrieval methods show significantly improved performance. This is because these additional queries often capture semantic nuances similar to those in the actual query \cite{nogueira2019document,nogueira2019doc2query,pradeep2021expando}. Importantly, \texttt{doc\-2query} shifts the computational load to the indexing phase, ensuring minimal performance lag during retrieval. By leveraging the \texttt{T5} model, the authors further enhanced the query generation quality, leading to the variation known as \texttt{doc\-TTTTTquery}, \texttt{doc\--T5query}, or \texttt{doc\-2query\--T5} \cite{nogueira2019doc2query}.
 
 \subsection*{\texttt{SPARTA} Sparse Retrieval}\label{sec:sparta}
-\texttt{SPARTA}, introduced by Zhao et al. \cite{zhao2020sparta}, represents a nuanced take on sparse retrieval. At its core, it works by encoding documents into sparse representations during the indexing phase. These representations not only capture the document's actual content but also incorporate terms that are semantically resonant, even if they're not present in the document. This underlying principle echoes the rationale of approaches like \texttt{doc2query} and dense retrieval models.
+\texttt{SPARTA}, introduced by Zhao et al. \cite{zhao2020sparta}, is an approach in the field of sparse information retrieval. At its core, it works by encoding documents into sparse representations during the indexing phase. These representations not only capture the document's actual content but also incorporate terms that are semantically resonant, even if they're not present in the document. This underlying principle echoes the rationale of approaches like \texttt{doc2query} and dense retrieval models.
 
-Yet, where \texttt{SPARTA} differentiates itself is in its retrieval phase. Unlike dense retrieval models, it retrieves pertinent documents using straightforward index lookups, mirroring lexical retrieval strategies like \texttt{BM25} \cite{zhao2020sparta}.
+Yet, where \texttt{SPARTA} differentiates itself is in its retrieval phase. Unlike dense retrieval models, it retrieves pertinent documents using straightforward index lookups, mirroring lexical retrieval strategies like \texttt{BM25} \cite{zhao2020sparta}. However, in real-world applications, \texttt{SPARTA} faces challenges. Several other models, including \texttt{BM25} and \texttt{doc2query-T5}, surpass it in ranking efficacy. Additionally, its indexing footprint is substantially larger compared to alternatives like \texttt{doc2query-T5} \cite{thakur2021beir}.
 
-However, in real-world applications, \texttt{SPARTA} faces challenges. Several other models, including \texttt{BM25} and \texttt{doc2query-T5}, surpass it in ranking efficacy. Additionally, its indexing footprint is substantially larger compared to alternatives like \texttt{doc2query-T5} \cite{thakur2021beir}.
+\subsection*{\texttt{ANCE} Dense Retrieval}\label{sec:ance}
+Dense retrieval is a newer approach in information retrieval. Traditional methods, like \texttt{BM25}, rely on specific term matching in high-dimensional spaces. These methods can sometimes miss the deeper meaning or context because they focus on explicit word matches. In contrast, dense retrieval uses continuous vector spaces to embed both queries and documents. In this approach, items that have similar meanings are closer together in the vector space, even if they don't share exact terms.
+
+\texttt{ANCE} (Approximate Nearest Neighbor Negative Contrastive Estimation) is an innovation in dense retrieval introduced by Xiong et al. \cite{xiong2020ance}. The main idea behind \texttt{ANCE} is to fix a problem in the training of dense retrieval models. Often, the irrelevant documents (or "negatives") used during training don't match the kind of irrelevant documents that come up during real-world testing. This mismatch can weaken the model's performance. To address this, \texttt{ANCE} updates its training data using an Approximate Nearest Neighbor (ANN) index. This index gets refreshed during the training process, making sure that the irrelevant documents chosen for training are similar to what the model will see in real-world scenarios. Thanks to this, \texttt{ANCE} retrieval pipelines perform better than those with other dense and sparse retrieval methods. In fact, \texttt{ANCE} achieves nearly the same accuracy as some sophisticated methods but is about 100 times faster \cite{xiong2020ance}.
 
 \subsection*{\texttt{monoT5} \& \texttt{duoT5} Rerankers}\label{sec:rerankers}
 \texttt{monoT5} and \texttt{duoT5} are neural re-rankers, also developed by Nogueira et al., which attempt to inject semantic understanding into the retrieval process \cite{nogueira2020document,nogueira2019multi}. Using the \texttt{T5} model, they re-rank a list of documents based on their semantic relevance to a given query. Specifically, \texttt{monoT5} processes a query and a single document, outputting a relevance score. In contrast, \texttt{duoT5} considers a query and two documents, determining which document is more relevant. Although \texttt{duoT5} offers a more nuanced ranking, its pairwise comparison method makes it computationally heavier. Hence, a staged re-ranking approach is proposed: first using \texttt{monoT5} for the top $k$ documents and subsequently applying \texttt{duoT5} to a smaller subset, the top $l$, where $l \ll k$ \cite{nogueira2019multi,pradeep2021expando}.
@@ -106,12 +109,14 @@ \subsection*{Expando-Mono-Duo Design Pattern}\label{sec:expando}
 The same research team introduced a strategic pattern for integrating the above tools into retrieval pipelines, termed the Expando-Mono-Duo design pattern \cite{pradeep2021expando}. Here's how it works: During indexing, \texttt{doc2query-T5} is employed to enhance document representation and better the initial retrieval results from methods like \texttt{BM25}. The retrieved results are then re-ranked with \texttt{monoT5}. A selected top tier from this list undergoes another re-ranking using \texttt{duoT5}. Trials show that this composite approach leads to marked improvements in result quality across multiple evaluation metrics \cite{pradeep2021expando}.
 
 \subsection*{Conversational Query Rewriting}\label{sec:cqr}
+Conversational search engines distinguish themselves from standard search engines by determining document relevance through the entirety of a conversation, not just the immediate query. In conversational contexts, subsequent questions often lean on prior interactions, implying that previous questions and answers must be factored in when fetching relevant documents. However, there's also a need to cater to conversation shifts where the immediate query doesn't relate to preceding exchanges. Blindly considering the entire conversational history in such cases could be detrimental to retrieval accuracy.
 
-Conversational search engines distinguish themselves from standard search engines by determining document relevance through the entirety of a conversation, not just the immediate query. In conversational contexts, subsequent questions often lean on prior interactions, implying that previous questions and answers must be factored in when fetching relevant documents. However, there's also a need to cater to conversation shifts where the immediate query doesn't relate to preceding exchanges. Blindly considering the entire conversational history in such cases could detriment retrieval accuracy.
+Elgohary et al. address this challenge with an innovative approach \cite{elgohary2019can}. They suggest reshaping the current query based on the overarching conversation. This reformulated query is designed to function autonomously within conventional retrieval pipelines. In essence, this technique extends the utility of standard search engines to conversational question-answering scenarios by introducing a preceding conversational query rewriting stage.
 
-Elgohary et al. address this challenge with an innovative approach \cite{elgohary2019can}. They suggest reshaping the current query based on the overarching conversation. This reformulated query is designed to function autonomously within conventional retrieval pipelines. In essence, this technique extends the utility of standard search engines to conversational question-answering scenarios by introducing a preceding conversational query modification stage.
+Employing text-to-text transformers, like \texttt{T5}, can be instrumental in achieving this rewrite. These models are trained to reformulate the user entered query, factoring in the conversational context. Studies validate the effectiveness of this approach, highlighting its capacity to enhance the retrieval accuracy of traditional search engines in conversational contexts \cite{elgohary2019can,anantha2020open,Lajewska:2023:ECIR}.
 
-Employing text-to-text transformers, like \texttt{T5}, can be instrumental in achieving this rewrite. These models are nurtured to revamp the immediate query, factoring in the conversational backdrop. Studies validate the efficacy of this approach, highlighting its capacity to enhance the retrieval accuracy of traditional search engines in conversational contexts \cite{elgohary2019can,anantha2020open,Lajewska:2023:ECIR}.
+\subsection*{\texttt{WaterlooClarke} Conversational Search Engine}\label{sec:waterlooclarke}
+In the third year of the TREC Conversational Assistance Track\footnote{URL: \url{https://www.treccast.ai/}}, the standout system was \texttt{WaterlooClarke}. It used a complex retrieval pipeline, as outlined in \cite{Lajewska:2023:ECIR,yan2021waterlooclarke}. The process starts with a \texttt{T5} component that rewrites the user's query. This step transforms a context-dependent question into one that's more general, free from the specific conversation's context, as outlined in Section \ref{sec:cqr}. Following this, the rewritten query is passed into two parallel retrieval stages: firstly, sparse retrieval through \texttt{BM25}, enhanced with pseudo-relevance feedback, and, secondly, dense retrieval powered by \texttt{ANCE}. To accommodate both approaches, \texttt{WaterlooClarke} employs two indices: a lexical one for \texttt{BM25} and an ANN index for \texttt{ANCE}. Once both retrieval stages conclude, their results are fused into a single list of ranked documents. This ranked list undergoes further refinement, first with \texttt{monoT5} and then a narrower re-ranking using \texttt{duoT5} for the top-tier documents \cite{Lajewska:2023:ECIR,yan2021waterlooclarke}.
 
 
 
@@ -152,7 +157,7 @@ \section{Incorporating Pseudo-Relevance Feedback into Our Baseline}\label{sec:ba
 
 Recognizing the substantial performance enhancements associated with pseudo-relevance feedback, we felt compelled to integrate a query expansion mechanism into our baseline retrieval method, see Section \ref{sec:baseline}. Our choice fell upon the \texttt{RM3} query expansion technique, well-established for its robustness and acceptance within the information retrieval community. For a deeper dive into its mechanics and principles, readers are directed to Section \ref{sec:prf}.
 
-In the \texttt{Pyterrier} framework, the setup requires that any query expansion follows an initial retrieval phase. This initial retrieval fetches the top $p$ documents, forming the foundation for subsequent query expansion using \texttt{RM3}. With the query expanded, it's then passed into a secondary retrieval phase to retrieve the final document set for the end-user. And, to fine-tune the output, we again apply re-ranking using both \texttt{monoT5} and \texttt{duoT5}.
+In the \texttt{Pyterrier} framework, the setup requires that any query expansion follows an initial retrieval phase. This initial retrieval fetches the top $p$ documents, forming the foundation for subsequent query expansion by $n$ words using \texttt{RM3}. With the query expanded, it's then passed into a secondary retrieval phase to retrieve the final document set for the end-user. And, to fine-tune the output, we again apply re-ranking using both \texttt{monoT5} and \texttt{duoT5}.
 
 Henceforth, we'll label this integrated retrieval approach as "baseline + \texttt{RM3}", which is structured as follows:
 \begin{enumerate}