[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-1-en-105":3,"doc-seo-253850-105":53,"doc-detail-253850-en":126},{"code":4,"msg":5,"data":6},0,"success",[7,14,19,24,29,34,39,44,49],{"id":8,"doc_module":9,"doc_module_name":10,"category_name":11,"show_sort_weight":12,"slug":13},11,1,"Template","Presentations",90,"presentations",{"id":15,"doc_module":9,"doc_module_name":10,"category_name":16,"show_sort_weight":17,"slug":18},12,"Resumes",80,"resumes",{"id":20,"doc_module":9,"doc_module_name":10,"category_name":21,"show_sort_weight":22,"slug":23},14,"Invoices",70,"invoices",{"id":25,"doc_module":9,"doc_module_name":10,"category_name":26,"show_sort_weight":27,"slug":28},15,"Posters",60,"posters",{"id":30,"doc_module":9,"doc_module_name":10,"category_name":31,"show_sort_weight":32,"slug":33},16,"Social Media",50,"social-media",{"id":35,"doc_module":9,"doc_module_name":10,"category_name":36,"show_sort_weight":37,"slug":38},17,"Forms",40,"forms",{"id":40,"doc_module":9,"doc_module_name":10,"category_name":41,"show_sort_weight":42,"slug":43},18,"Letters",30,"letters",{"id":45,"doc_module":9,"doc_module_name":10,"category_name":46,"show_sort_weight":47,"slug":48},21,"Paper Templates",5,"papers-templates",{"id":50,"doc_module":9,"doc_module_name":10,"category_name":51,"show_sort_weight":4,"slug":52},158,"General","general-158",{"code":4,"msg":54,"data":55},"ok",{"site_id":56,"language":57,"slug":58,"title":59,"keywords":60,"description":61,"schema_data":62,"social_meta":119,"head_meta":121,"extra_data":123,"updated_unix":125},105,"en","improving-nhl-draft-outcome-predictions-using-scouting-reports","Improving NHL Draft Outcome Predictions using Scouting Reports","","We leverage Large Language Models (LLMs) to extract structured information from NHL scouting report texts and enhance predictions of NHL draft outcomes. In parallel, the work derives statistical features from players’ on-ice performance leading up to the draft. These two feature sets are combined with ensemble machine learning, where both modalities contribute predictive signal. The strongest results emerge when scouting-report-derived features and on-ice statistics are jointly modeled.",{"@graph":63,"@context":118},[64,80,101],{"@type":65,"itemListElement":66},"BreadcrumbList",[67,71,74,77],{"item":68,"name":69,"@type":70,"position":9},"https://docshare.wps.com","Home","ListItem",{"item":72,"name":10,"@type":70,"position":73},"https://docshare.wps.com/template/",2,{"item":75,"name":11,"@type":70,"position":76},"https://docshare.wps.com/template/presentations/",3,{"item":78,"name":59,"@type":70,"position":79},"https://docshare.wps.com/template/improving-nhl-draft-outcome-predictions-using-scouting-reports/253850/",4,{"url":78,"name":59,"@type":81,"image":82,"author":87,"headline":59,"publisher":90,"fileFormat":93,"inLanguage":57,"description":61,"dateModified":94,"datePublished":95,"encodingFormat":93,"isAccessibleForFree":96,"interactionStatistic":97},"DigitalDocument",{"url":83,"@type":84,"width":85,"height":86},"https://docshare.wps.com/thumbnails/improving-nhl-draft-outcome-predictions-using-scouting-reports/253850.png","ImageObject",442,249,{"name":88,"@type":89},"Berry Peter","Person",{"url":68,"name":91,"@type":92},"DocShare","Organization","application/pdf","2026-10-31","2026-09-13",true,{"@type":98,"interactionType":99,"userInteractionCount":20},"InteractionCounter",{"@type":100},"ViewAction",{"@type":102,"mainEntity":103},"FAQPage",[104,110,114],{"name":105,"@type":106,"acceptedAnswer":107},"What data sources does the paper use to predict NHL draft outcomes?","Question",{"text":108,"@type":109},"It uses scouting report text extracted via LLMs and statistical features derived from a player’s on-ice performance prior to the draft.","Answer",{"name":111,"@type":106,"acceptedAnswer":112},"How are the scouting-report and on-ice features combined in the modeling?",{"text":113,"@type":109},"The paper trains base models independently on each dataset and then combines them using a stacking ensemble approach.",{"name":115,"@type":106,"acceptedAnswer":116},"What is the main finding about using scouting reports versus on-ice statistics?",{"text":117,"@type":109},"Both on-ice statistics and scouting reports are predictive, and combining them yields the strongest prediction performance.","https://schema.org",{"og:url":78,"og:type":120,"og:title":59,"og:site_name":91,"og:description":61},"article",{"robots":122,"canonical":78},"index,follow",{"doc_id":124,"site_id":56},253850,1791475777,{"code":4,"msg":5,"data":127},{"doc_id":124,"user_id":128,"nickname":88,"user_avatar":129,"doc_module":9,"category_id":8,"category_name":11,"doc_title":59,"doc_description":61,"doc_content":130,"file_id":131,"file_url":132,"file_type":133,"file_size":134,"view_count":20,"is_deleted":4,"is_public":9,"is_downloadable":9,"audit_status":9,"page_count":135,"language":136,"language_code":57,"site_id":56,"html_lang":57,"table_of_contents":137,"faqs":138,"seo_title":139,"seo_description":61,"update_tm":140,"read_time":141},1374402524268,"https://ap-avatar.wpscdn.com/davatar_3d24733baf745e90a7e4bdd5f77d97b2","Improving NHL Draft Outcome Predictions using Scouting Reports  \nHubert Luo. 1  \nAbstract  \nWe leverage Large Language Models (LLMs) to extract information from scouting report texts and improve predictions of National Hockey League (NHL) draft outcomes. In parallel, we derive statistical features based on a player’s on-ice performance leading up to the draft. These two datasets are then combined using ensemble machine learning models. We find that both on-ice statistics and scouting reports have predictive value, however combining them leads to the strongest results.  \n1. Introduction  \nHockey has long emphasized the importance of the eye test: evaluating players using visual observations of hockey games. However, the eye test is prone to biases (Deaneret al., 2013), anecdotal evidence, and the reality that machines can incorporate information from a much larger sample size of games than a human can.  \nOn the other hand, analytical models lack situational context, nuance, and information that only humans can collect from discussions with other humans (Wolfson et al., 2011) . Therefore, this paper’s objective is to leverage the strengths of both approaches by extracting information gleaned from the eye test and combining it with quantitative analyses.  \nLeading up to the NHL draft, scouts attend games worldwide to identify top draft-eligible players, write scouting reports for them, and compare prospects with their peers through rankings. These scouting report texts are a rich source of qualitative information about players, encompassing information about their playing style, strengths/weaknesses, and a scout’s feedback on how well a prospect’s ability will translate to the professional game.  \nFor example, scout Scott Wheeler writes about Connor Bedard, who ended up being drafted first overall in the 2023 NHL Draft:  \n ”[Bedard]’s got unbelievably quick hands and  \n1Department of Computer Science, University of Texas at Austin and Data Analytics Group, Lazard.  \nCopyright 2024 by the author.  \nthe loose grip that all great handlers have. He’s got high-end speed with his galloping crossover strides and strong acceleration from a standstill, which help him carry the puck up the ice, create one-on-one off the rush, or join in transition as the trailer whenever he has to play catchup.”  \n(Wheeler, 2023a)  \nScouts spend a vast amount of time canvassing thousands of potential prospects, providing unique information and learning important context about a player’s on-ice performance that is only observable by a person or by interviewing coaches, prospects, and their teammates. This off-ice context is an important factor in a player’s performance that has typically only been available to humans.  \nIncorporating these scouting reports into existing hockey analytics processes has been particularly challenging given the need to translate these texts into usable and comprehensive quantitative features. This is an area where LLMs have demonstrated strong potential over the last year, for example in engineering features for financial stock predictions (Lopez-Lira and Tang, 2023) . This paper will be one of the first to apply LLMs to hockey, allowing us to integrate off-ice context into analytical models using scouting reports.  \nWe end up with two different datasets, one featuring a player’s on-ice statistics prior to the draft, and another with features engineered from a player’s scouting reports. Base models are trained independently on each dataset in parallel, and then combined using a stacking ensemble learning methodology. Stacking was chosen to mitigate spurious decision-making and facilitate better comparison between base models as outlined in section 3.10.  \nIn this paper, we will demonstrate that LLMs are useful in extracting information from scouting reports and that combining it with on-ice statistics will improve our predictions of post-draft outcomes.  \n2. Previous Work  \nRecent years have seen an uptick in the adaptation of s","cbCaiaefxpqpyh4b","https://ap.wps.com/l/cbCaiaefxpqpyh4b","pdf",360627,19,"English","# Abstract\n# Introduction\n# Previous Work","[{\"question\":\"What data sources does the paper use to predict NHL draft outcomes?\",\"answer\":\"It uses scouting report text extracted via LLMs and statistical features derived from a player’s on-ice performance prior to the draft.\"},{\"question\":\"How are the scouting-report and on-ice features combined in the modeling?\",\"answer\":\"The paper trains base models independently on each dataset and then combines them using a stacking ensemble approach.\"},{\"question\":\"What is the main finding about using scouting reports versus on-ice statistics?\",\"answer\":\"Both on-ice statistics and scouting reports are predictive, and combining them yields the strongest prediction performance.\"}]","Improving NHL Draft Outcome Predictions using Scouting Reports | PDF",1789270991,7]