Analysis and Advancement in Domain-Specific Templated Question Answering

<!DOCTYPE html> <html> <head>  <link href="https://fonts.googleapis.com/icon?family=Material+Icons" rel="stylesheet"> <link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.0.13/css/all.css" integrity="sha384-DNOHZ68U8hZfKXOrtjWvjxusGo9WQnrNx2sqG0tfsghAvtVlRW3tvkXWZh58N9jp" crossorigin="anonymous"> <link href="https://fonts.googleapis.com/css?family=Roboto" rel="stylesheet">  <link type="text/css" rel="stylesheet" href="css/materialize.min.css" media="screen,projection" /> <link type="text/css" rel="stylesheet" href="css/main.css" />  <title>Analysis and Advancement in Domain-Specific Templated Question Answering</title>  <meta name="viewport" content="width=device-width, initial-scale=1.0" /> <meta name="title" content="Analysis and Advancement in Domain-Specific Templated Question Answering"> <meta name="description" content="This work addresses the challenge of domain-specific question answering through the intelligent composition of tool sequences using a large language model. We formulate the problem as utilizing a set of tools T to answer a query Q by determining the necessary tools, arguments, and execution sequence. Our approach enhances language model capabilities through prompt engineering, leveraging advanced reasoning, and adopting our custom Chain of Thoughts (CoT) inspired strategy for dynamic, cascaded user engagement. Employing multi-task learning broadens knowledge scope, while transfer learning from domains with richer tooling enhances versatility. Runtime compute costs are optimized through distillation. The evaluation shows our method excels in selecting optimal tool combinations for domain-specific queries, outperforming baseline approaches in accuracy and coverage. This approach provides a reusable framework for constructing proficient and cost-effective domain-specific Question Answering (QA) solutions. Key explorations encompass analysis of prompt engineering for tool interfaces, compositional learning across tools, transfer learning from richer domains, and prompt distillation. These facilitate the practical deployment of LLMs for industrial applications. "> <meta name="keywords" content="Query, Tool, Tool Retrieval, Chain of Thoughts(CoT) Prompting, Prompt Engineering, QA, Distillation Step by Step, Array of Thoughts(AoT), GPT, LLM, Rationale ., Computer Science, Technology, open access proceedings"/>   <meta name="dc.title" content="Analysis and Advancement in Domain-Specific Templated Question Answering"> <meta name="citation_authors" content="Aaditya Baranwal"> <meta name="citation_authors" content="Jyotin Goel"> <meta name="citation_authors" content="Prashant Tandon"> <meta name="citation_authors" content="Renu Sankhla"> <meta name="citation_authors" content="Sukriti Goyal"> <meta name="dc.type" content="Article"> <meta name="dc.source" content="Computer Science & Information Technology (CS & IT) Vol.14, No.08"> <meta name="dc.date" content="2024/04/28"> <meta name="dc.identifier" content="10.5121/csit.2024.140818"> <meta name="dc.publisher" content="AIRCC Publishing Corporation"> <meta name="dc.rights" content="http://creativecommons.org/licenses/by/3.0/"> <meta name="dc.format" content="application/pdf"> <meta name="dc.language" content="en"> <meta name="dc.description" content="This work addresses the challenge of domain-specific question answering through the intelligent composition of tool sequences using a large language model. We formulate the problem as utilizing a set of tools T to answer a query Q by determining the necessary tools, arguments, and execution sequence. Our approach enhances language model capabilities through prompt engineering,leveraging advanced reasoning, and adopting our custom Chain of Thoughts (CoT)inspired strategy for dynamic, cascaded user engagement. Employing multi-task learning broadens knowledge scope, while transfer learning from domains with richer tooling enhances versatility. Runtime compute costs are optimized through distillation. The evaluation shows our method excels in selecting optimal tool combinations for domainspecific queries, outperforming baseline approaches in accuracy and coverage. This approach provides a reusable framework for constructing proficient and cost-effective domain-specific Question Answering (QA) solutions. Key explorations encompass analysis of prompt engineering for tool interfaces, compositional learning across tools, transfer learning from richer domains, and prompt distillation. These facilitate the practical deployment of LLMs for industrial applications."/> <meta name="dc.subject" content="Query"> <meta name="dc.subject" content="Tool"> <meta name="dc.subject" content="Tool Retrieval"> <meta name="dc.subject" content="Chain of Thoughts(CoT) Prompting"> <meta name="dc.subject" content="Prompt Engineering, QA"> <meta name="dc.subject" content="Distillation Step by Step"> <meta name="dc.subject" content="Array of Thoughts(AoT)"> <meta name="dc.subject" content="GPT"> <meta name="dc.subject" content="LLM"> <meta name="dc.subject" content="Rationale">   <meta name="prism.publicationName" content="Computer Science & Information Technology (CS & IT)"> <meta name="prism.publicationDate" content="2024/04/28"> <meta name="prism.volume" content="14"> <meta name="prism.number" content="08"> <meta name="prism.section" content="Article"> <meta name="prism.startingPage" content="185">   <meta name="citation_journal_title" content="Computer Science & Information Technology (CS & IT)"> <meta name="citation_publisher" content="AIRCC Publishing Corporation"> <meta name="citation_authors" content="Aaditya Baranwal, Jyotin Goel, Prashant Tandon,Renu Sankhla and Sukriti Goyal"> <meta name="citation_title" content="Analysis and Advancement in Domain-Specific Templated Question Answering"> <meta name="citation_online_date" content="2024/04/28"> <meta name="citation_issue" content="14"> <meta name="citation_firstpage" content="185"> <meta name="citation_authors" content="Aaditya Baranwal"> <meta name="citation_authors" content="Jyotin Goel"> <meta name="citation_authors" content="Prashant Tandon"> <meta name="citation_authors" content="Renu Sankhla"> <meta name="citation_authors" content="Sukriti Goyal"> <meta name="citation_doi" content="10.5121/csit.2024.140818"> <meta name="citation_abstract_html_url" content="https://aircconline.com/csit/abstract/v14n8/csit140818.html"> <meta name="citation_pdf_url" content="https://aircconline.com/csit/papers/vol14/csit140818.pdf">   <meta property="og:site_name" content="AIRCC" /> <meta property="og:type" content="article" /> <meta property="og:url" content="https://aircconline.com/csit/abstract/v14n8/csit140818.html"> <meta property="og:title" content="Analysis and Advancement in Domain-Specific Templated Question Answering"> <meta property="og:description" content="This work addresses the challenge of domain-specific question answering through the intelligent composition of tool sequences using a large language model. We formulate the problem as utilizing a set of tools T to answer a query Q by determining the necessary tools, arguments, and execution sequence. Our approach enhances language model capabilities through prompt engineering,leveraging advanced reasoning, and adopting our custom Chain of Thoughts (CoT)inspired strategy for dynamic, cascaded user engagement. Employing multi-task learning broadens knowledge scope, while transfer learning from domains with richer tooling enhances versatility. Runtime compute costs are optimized through distillation. The evaluation shows our method excels in selecting optimal tool combinations for domainspecific queries, outperforming baseline approaches in accuracy and coverage. This approach provides a reusable framework for constructing proficient and cost-effective domain-specific Question Answering (QA) solutions. Key explorations encompass analysis of prompt engineering for tool interfaces, compositional learning across tools, transfer learning from richer domains, and prompt distillation. These facilitate the practical deployment of LLMs for industrial applications."/>   <meta name="twitter:card" content="Proceedings" /> <meta name="twitter:site" content="AIRCC" /> <meta name="twitter:title" content="Analysis and Advancement in Domain-Specific Templated Question Answering" /> <meta name="twitter:description" content="This work addresses the challenge of domain-specific question answering through the intelligent composition of tool sequences using a large language model. We formulate the problem as utilizing a set of tools T to answer a query Q by determining the necessary tools, arguments, and execution sequence. Our approach enhances language model capabilities through prompt engineering,leveraging advanced reasoning, and adopting our custom Chain of Thoughts (CoT)inspired strategy for dynamic, cascaded user engagement. Employing multi-task learning broadens knowledge scope, while transfer learning from domains with richer tooling enhances versatility. Runtime compute costs are optimized through distillation. The evaluation shows our method excels in selecting optimal tool combinations for domainspecific queries, outperforming baseline approaches in accuracy and coverage. This approach provides a reusable framework for constructing proficient and cost-effective domain-specific Question Answering (QA) solutions. Key explorations encompass analysis of prompt engineering for tool interfaces, compositional learning across tools, transfer learning from richer domains, and prompt distillation. These facilitate the practical deployment of LLMs for industrial applications."/> <meta name="twitter:image" content="https://airccse.org/img/aircc-logo1.jpg" />  <style type="text/css"> .rdd { text-align: center; background: #f2f2f2; color: #000; font-weight: 700; width: 130px; height: 110px; border-radius: 100%; box-shadow: inset 1px 0px 22px 3px #4080ca; font-family: 'Oswald', sans-serif; border: 5px solid #1f8ea3; margin: 5% auto; line-height: 110px; } </style> <script async src="//pagead2.googlesyndication.com/pagead/js/adsbygoogle.js"></script> <script> (adsbygoogle = window.adsbygoogle || []).push({ google_ad_client: "ca-pub-1537319084895272", enable_page_level_ads: true }); </script> </head> <body>  <div class="navbar-fixed"> <nav class="cyan lighten-2 z-depth-5"> <div class="container"> <div class="nav-wrapper"> <ul> <li id="b-logo"> <img id="brand-logo" href="index.php" class="hide-on-med-and-down" src="img/aircc-logo1.jpg"> </li> </ul> <a class="brand-logo" href="index.php">AIRCC</a> <a data-activates="side-nav" class="button-collapse show-on-small left"> <i class="material-icons">menu</i> </a> <ul class="right hide-on-med-and-down"> <li > <a href="https://aircconline.com/">Home</a> </li> <li> <a href="https://airccse.org/csit/V14N08.html">Current Issue</a> </li> <li> <a href="https://airccse.org/arch.html">Archives</a> </li> <li> <a href="https://airccse.org/csit/acontact.html">Contact</a> </li> <li> <a class="openIcon" onclick="openSearch()"> <i class="material-icons">search</i> </a> </li> </ul> </div> </div> </nav> </div>  <ul class="side-nav" id="side-nav"> <li> <div class="user-view arc"> <div class="background"> <img class="mobile-overlay" > </div> <a href="https://aircconline.com/"> <i id="cl" class="material-icons cyan-text text-lighten-2 right">close</i> </a> <a href="https://aircconline.com/"> <img class="circle" src="img/aircc-logo1.jpg"> </a> <h4 class="grey-text">AIRCC</h4> </div> </li> <li > <a href="https://aircconline.com/">Home <i class="material-icons">home</i> </a> </li> <li> <a href="https://airccse.org/csit/V14N08.html">Current Issue <i class="fas fa-users"></i> </a> </li> <li> <a href="https://airccse.org/arch.html">Archives <i class="fas fa-users"></i> </a> </li> <li> <a href="https://airccse.org/csit/acontact.html">Contact <i class="fas fa-calendar-alt"></i> </a> </li> <li>  <li> <a class="openIcon" id="icon" onclick="openSearch()"> <i class="material-icons">search</i> Search </a> </li> </ul>  <div id="myOverlay" class="overlay"> <span class="closeIcon" onclick="closeSearch()" title="Close Overlay">×</span> <div class="overlay-content"> <form action="https://airccj.org/csecfp/library/index.php"> <input type="text" placeholder="Search.." name="title"> <button type="submit"> <i class="material-icons center">search</i> </button> </form> </div> </div>  <section class="section-main"> <div class="container"> <div class="row"> <div class="col s12 m8"> <div class="card z-depth-2"> <div class="card-content"> <h5 class="cyan-text center text-darken-1"> Analysis and Advancement in Domain-Specific Templated Question Answering </h5> </div> </div> <br> <div class="card"> <h5 id="about" class="brown-text text-darken-2 text-center" style="padding-bottom:0px">Authors</h5>  <div class="card-content"> <p class="left-text" style="text-align:justify"> Aaditya Baranwal, Jyotin Goel, Prashant Tandon, Renu Sankhla and Sukriti Goyal, Indian Institute of Technology Jodhpur, India </p> </div> </div>   <div class="card"> <h5 id="about" class="brown-text text-darken-2 text-center" style="padding-bottom:0px">Abstract</h5>  <div class="card-content"> <p class="left-text" style="text-align:justify"> This work addresses the challenge of domain-specific question answering through the intelligent composition of tool sequences using a large language model. We formulate the problem as utilizing a set of tools T to answer a query Q by determining the necessary tools, arguments, and execution sequence. <br> Our approach enhances language model capabilities through prompt engineering, leveraging advanced reasoning, and adopting our custom Chain of Thoughts (CoT) inspired strategy for dynamic, cascaded user engagement. Employing multi-task learning broadens knowledge scope, while transfer learning from domains with richer tooling enhances versatility. Runtime compute costs are optimized through distillation. The evaluation shows our method excels in selecting optimal tool combinations for domain-specific queries, outperforming baseline approaches in accuracy and coverage. This approach provides a reusable framework for constructing proficient and cost-effective domain-specific Question Answering (QA) solutions. <br> Key explorations encompass analysis of prompt engineering for tool interfaces, compositional learning across tools, transfer learning from richer domains, and prompt distillation. These facilitate the practical deployment of LLMs for industrial applications. </p> </div> </div> <div class="card"> <h5 id="about" class="brown-text text-darken-2 text-center" style="padding-bottom:0px">Keywords</h5>  <div class="card-content"> Query, Tool, Tool Retrieval, Chain of Thoughts(CoT) Prompting, Prompt Engineering, QA, Distillation Step by Step, Array of Thoughts(AoT), GPT, LLM, Rationale . </p> </div> </div> <div class="card-content"> <a href="https://aircconline.com/csit/papers/vol14/csit140818.pdf" target="_blank" class="btn btn-small lighten-2 cyan lig">Full Text</a>  <a href="https://airccse.org/csit/V14N08.html" target="_blank" class="btn btn-small lighten-2 cyan lig">Volume 14 Number 08</a> </div> </div>  <div id="side-bar" class="col s12 m4"> <div id="section-main"> <div class="card side cyan lighten-2"> <div class="card-content"> <ul> <li class="ax waves-effect waves-light"> <a class="white-text" href="https://airccse.org/editorial.html" target="blank"><i class="material-icons left">account_circle</i>Editorial Board</a> <br> </li> <br> <br> <div class="divider"></div> <br> <li class="ax waves-effect waves-light"> <a class="white-text" href="https://airccse.org/arch.html" target="blank"> <i class="material-icons fa fa-archive left"></i>Archives </a> <br> </li> <br> <br> <div class="divider"></div> <br> <li class="ax waves-effect waves-light"> <a class="white-text" target="blank" href="https://airccse.org/indexing.html"> <i class="material-icons left">local_pharmacy</i>Indexing</a> </li> <br> <br> <div class="divider"></div> <br> <li class="ax waves-effect waves-light"> <a class="white-text" target="blank" href="http://airccse.org/faq.html"> <i class="material-icons left">quiz</i>FAQ</a> </li> </ul> </div> </div> </div> </div> </div> </div> </section>  <div id="txtcnt"></div>  <footer class="page-footer cyan lighten-3"> <div class="container"> <div class="row"> <div class="footer-m col s12 m6 l3 "> <ul> <li> <img src="img/since2008.png" alt="since2008"> </li> </ul> </div> <div class="footer-m col s12 m6 l3 "> <ul> <li> <a class="white-text" href="ethics.html">Ethics</a> </li> <li> <a class="white-text" href="faq.html">FAQ</a> </li> <li> <a class="white-text" href="subscription.html">Subscription</a> </li> </ul> </div> <div class="footer-m col s12 m6 l3 offset-m1"> <ul> <li> <a class="white-text" href="acontact.html">Contact</a> </li> <li> <a class="white-text" href="https://airccse.org/sitemap.html">Sitemap</a> </li> </ul> </div> <div class="social col s12 m6 l3 offset-m1"> <ul> <li> <a class="blue-text text-darken-4" href="https://www.facebook.com/AIRCCPC" target="blank"> <i class="fab fa-facebook"> </i> </a> </li> <li> <a class="cyan-text " href="https://twitter.com/AIRCCFP" target="blank"> <i class="fab fa-twitter"></i> </a> </li> <li> <a class="red-text text-darken-4" href="https://www.youtube.com/channel/UCzkuYvuKuNCIc3jbE52IeZg" target="blank"> <i class="fab fa-youtube"></i> </a> </li> </ul> </div> </div> </div> <div class="footer-copyright grey darken-2"> <div class="container center-align"> <large class="white-text">Not for Profit @ All Rights Reserved ® AIRCC </large> </div> </div>  <div class="col s12 m10 offset-m1"> <div class="grey darken-3 center-align"> <small class="white-text">Designed and Developed by Wireilla Delivery Team</small> </div> </div> </footer> <script type="text/javascript" src="https://code.jquery.com/jquery-3.2.1.min.js"></script> <script type="text/javascript" src="js/materialize.min.js"></script> <script src="js/scrolltop.js"></script> <script src="js/search.js"></script> <script src="js/popup.js"></script> <script src="js/main.jquery.js"></script> </body>  </html>

CINXE.COM

Analysis and Advancement in Domain-Specific Templated Question Answering