webis/tip-of-my-tongue-known-item-search-triplets

The TOMT-KIS-TRIPLETS (tip-of-my-tongue-known-item-search triplets) Dataset

TOMT-KIS-TRIPLETS is a refined subset of the broader TOMT-KIS dataset, curated for advanced applications. Given that Wikipedia and IMDb are besides YouTube the most commonly appearing domains in links on the answer path (also see in TOMT-KIS the column links_on_answer_path), we focus specifically on these sources for higher relevancy and usability. By leveraging a Wikipedia dump and SPARQL Wikipedia Query Service, our objective is to develop a labeled dataset that contains direct links to relevant Wikipedia articles, making it ideal for supervised learning tasks.

Citation

If you use the TOMT-KIS triplets, please cite the corresponding paper:

@InProceedings{froebe:2023c,
  author =                   {Maik Fr{\"o}be and Eric Oliver Schmidt and Matthias Hagen},
  booktitle =                {QPP++ 2023: Query Performance Prediction and Its Evaluation in New Tasks},
  month =                    apr,
  publisher =                {CEUR-WS.org},
  series =                   {CEUR Workshop Proceedings},
  site =                     {Dublin, Irland},
  title =                    {{A Large-Scale Dataset for Known-Item Question Performance Prediction}},
  year =                     2023
}

Use of TOMT-KIS-TRIPLETS

You can download our full dataset using the datasets library.

from datasets import load_dataset
tomt_kis_triplets = load_dataset("webis/tip-of-my-tongue-known-item-search-triplets")

To see the size of the dataset execute

len(tomt_kis_triplets['train'])

You can access the first respectively last row in TOMT-KIS using

# Get the first row in the dataset
tomt_kis_triplets['train'][0]


# Get the last row in the dataset
tomt_kis_triplets['train'][-1]

... and a specific column:

# Returns a list of all the values in the the title column
tomt_kis_triplets['train']['title']

# Returns the title of the first column
tomt_kis_triplets['train'][0]['title']

Iterate over the TOMT-KIS dataset:

for element in tomt_kis_triplets['train']:
    print(element)

For further information see the Datasets documentation.

Dataset Structure

Our TOMT-KIS-TRIPLETS dataset is available in JSONL format. Important to mention is that the qid is identical to id in TOMT-KIS and therefore the last mentioned can be joined easily in order to access all of its 128 attributes.

Data Instances

{
  "qid": "ohefyy",
  "query": "Assassins pitted against each other\n\nWatched it back in high school. Particular scene I remember was a priest got caught up in the fights because he accidentally swallowed a small exploding device which was put by an assassin in his coffee. Each assassin had it in them, and there was some sort of time limit if I'm not mistaken.",
  "docno_positive": 12054679,
  "url_wikipedia_positive": "https://en.wikipedia.org/wiki/The Tournament (2009 film)",
  "positive": "The Tournament (2009 film)\n\nThe Tournament is a 2009 British independent action thriller film, marking the directorial debut of Scott Mann. The film was conceived by Jonathan Frank and Nick Rowntree while at the University of Teesside with Mann. The script was written by Gary Young, Jonathan Frank, and Nick Rowntree.\n\nThe Tournament was partially filmed in Bulgaria, and numerous locations around Northern England (where the film is set) and Merseyside. The film stars Robert Carlyle, Ving Rhames, Kelly Hu, Sébastien Foucan, Liam Cunningham, Scott Adkins, Camilla Power, and Ian Somerhalder. The film received additional funding internationally, from Sherezade Film Development, Storitel Production, and others, earning the film a budget of just under £4,000,000. The film also features a renowned international ensemble cast.\n\nHowever, numerous problems involving production, finance (the budget ran out twice), and securing a distributor meant the film was not released until two years after filming, in late 2009.\n\nPlot\nA group of the world's wealthiest and most powerful men, involved with law enforcement agencies and criminal underworlds alike, control the international media with their billions. They devise the ultimate entertainment, known as \"The Tournament\", which takes place every seven or ten years in an unspecified large city.\n\n\"Contestants\" volunteer for this event, often the toughest or craziest individuals on the planet. The last surviving combatant wins a cash prize of $10 million. The tournament organizers not only view this as entertainment but also wager high stakes on the outcome, creating a major gambling tournament.\n\nCombatants are selected from the world's special forces, serial killers, athletes, trained contract killers, and assassins. Each carries a tracking device embedded under their skin, allowing organizers to monitor their movements. The tournament lasts 24 hours, and if no one has won by the end, then...",
  "negative": "Metalocalypse: The Doomstar Requiem\n\nMetalocalypse: The Doomstar Requiem - A Klok Opera is a rock opera special for the Adult Swim animated series \"Metalocalypse\". Announced on May 10, 2013, by Adult Swim, it was released on October 27, 2013. The special picks up directly after \"Metalocalypse\" season 4. It features all-new music from Dethklok, released on a soundtrack album on October 29, 2013.\n\nPlot\nThe special begins with Ishnifus Meaddle chanting with the monks of the Church of the Black Klok about the Doomstar's arrival and the prophecy that one of them must die before the \"prophet's night\" is over, as Charles Offdensen looks on with concern over the impending \"Metalocalypse\" (\"The Birth/Fata Sidus Oritur/One of Us Must Die\"). Magnus Hammersmith and the Metal Masked Assassin hold Toki Wartooth and Abigail Remeltindrinc prisoner to lure Dethklok into their trap (\"Magnus and the Assassin\"). Meanwhile, Dethklok \"parties around the world\" to forget Toki's disappearance (\"Partying Around the World\"), while Offdensen attempts to locate Toki. The band's best operative returns dead, with a USB flash drive found in his remains. It contains a video of Magnus holding Toki and Abigail hostage, revealing his location in the \"depths of humanity.\"\n\nIshnifus urges Dethklok to rescue Toki, but the band refuses his plea (\"How Can I Be A Hero?\"). During a live show, a malfunctioning hologram of Toki leads fans to boo Dethklok. Realizing they have no other choice, Dethklok decides to rescue Toki (\"The Fans Are Chatting\").\n\nAbigail comforts Toki, urging him to find solace in his \"happy place\" (\"Abigail's Lullaby\"). This recalls the day Toki first auditioned for Dethklok, replacing the disgraced Magnus. After a fierce guitar battle with Skwisgaar Skwigelf, Toki joins the band...",
  "docno_negative": 39371831,
  "categories": [ "[TOMT]", "[MOVIE]" ]
}

Data Fields

The TOMT-KIS-TRIPLETS dataset includes the following key columns:

• qid (string): Query ID from Reddit. Identical to id in TOMT-KIS.
• query (string): The full Reddit question, combining the Title with the Content, separated by '\n\n'. Note that the selected Reddit answer as well as the category tags are excluded from this field.
• docno_positive (int): Document ID of the positive document within the Wikipedia dump, associated with the query.
• url_wikipedia_positive (string): Wikipedia URL of the positive document linked to the query.
• positive (string): The content of the Wikipedia article serving as the positive document for the query.
• negative (string): The content of the Wikipedia article serving as the negative document for the query.
• docno_negative (int): Document ID for the negative document within the TREC-ToT 2024 corpus.
• categories (list of strings): list of all category tags that were part of the original title (query)

Data processing steps

Our target is to provide a dataset with a labeled Wikipedia URL per solved question. We filter our TOMT-KIS dataset for questions with a chosen_answer that includes links to Wikipedia or IMDb. Since the chosen_answer links aren’t explicitly provided, we first extract URLs from the chosen answers. Then, we filter these URLs to include only those with Wikipedia or IMDb as the domain. Canonicalization of URLs is performed to ensure consistency and accurate matching across sources. We used the SPARQL service to retrieve the corresponding wikidata_urls for each IMDb URL we extracted during this process in the chosen answer to a question. The filtered dataset includes 32,553 Reddit questions with a Wikipedia link, successfully matched to the Wikipedia dump.

In order to ensure training data without bias, we remove all entries from TOMT-KIS-Triplet that either a) contain a url_wikipedia_positive of the positive document of an entry that occurs or b) contain a docno_negative of the negative document of an entry that occurs in on of the qrels files from TREC-ToT 2024.

• negative_trec_tot_2024_id can directly be used to filter out the corresponding entries. This way, 240 entries where filtered out of our dataset.
• As TREC-ToT 2024 only contains wikidata_ids of each document, we first need to use SPARQL in order to retrieve the corresponding wikidata_urls for each wikidata_id in each qrelss file before filtering. This way, 58 entries where filtered out of our dataset.
• In total, TOMT-KIS-Triplet contains 32,553 entries.