* Updated JSON files at this other Kaggle repository
- This dataset has been produced as part of the The Web of Know-How project
- To cite this dataset use: Paolo Pareti, Benoit Testu, Ryutaro Ichise, Ewan Klein and Adam Barker. Integrating Know-How into the Linked Data Cloud. Knowledge Engineering and Knowledge Management, volume 8876 of Lecture Notes in Computer Science, pages 385-396. Springer International Publishing (2014) (PDF) (bibtex)
- Quickstart: if you want to experiment with the most high-quality data before downloading all the datasets, download the file 9of11_knowhow_wikihow, and optionally files instruction set entities, Process - Inputs, Process - Outputs, Process - Step Links and wikiHow categories hierarchy.
- Data representation based on the PROHOW vocabulary
- Data extracted from existing web resources is linked to the original resources using the Open Annotation specification
- Data concerning the manual evaluation of this dataset is available here
- Data also available from datahub and kaggle
Instruction datasets:
- Datasets 1of11_knowhow_wikihow to 9of11_knowhow_wikihow contain instructions from the wikiHow website. Instructions are allocated in the datasets in order of popularity. This means that the most popular and high-quality instructions are found in 9of11_knowhow_wikihow, while the least popular ones are in dataset 1of11_knowhow_wikihow. These instructions are also classified according to the hierarchy found in wikiHow categories hierarchy. wikiHow instructions are community generated. As a result of this, each task is described by at most one set of instructions, which is usually of high quality thanks to the collective community contributions.
- Datasets 10of11_knowhow_snapguide to 11of11_knowhow_snapguide contain instructions from the Snapguide website. These instructions are not sorted by their popularity, but they are classified in one of the Snapguide categories. Snapguide instructions are created by single users. As a result of this there might be multiple sets of instructions to achieve the same task; however these instructions on average contain more noise as they are not peer-reviewed.
Links datasets:
- The Process - Inputs datasets contain detailed information about the inputs of the sets of instructions, including links to DBpedia resources
- The Process - Outputs datasets contains detailed information about the outputs of the sets of instructions, including links to DBpedia resources
- The Process - Step Links datasets contains links between different sets of instructions
Other datasets:
- The wikiHow categories hierarchy dataset contains information on how the various wikiHow categories are hierarchically structured, and how they relate to the Snapguide categories.
- The instruction set entities dataset lists all the top-level entities in a sets of instructions. In other words, all the entities which correspond to the title of a set of instructions.
- The wikiHow community links dataset lists the links manually created by the wikiHow community of users that interlink entities belonging to different sets of instructions.
The following figure is a simple example of how the PROHOW vocabulary is used in the datasets. Instructions in the dataset can have more complex structures, for example instructions could have multiple methods, steps could have further sub-steps, and complex requirements could be decomposed into sub-requirements.
Sample SPARQL queries are available here.
A sample SPARQL endpoint is available at Dydra. You can use this URI for federated queries. To get you started, here are some sample queries. NOTE: this endpoint exposes only a subset of the dataset, more specifically files:
- 8of11_knowhow_wikihow
- 9of11_knowhow_wikihow
- instruction set entities
- Process - Inputs
- Process - Outputs
- Process - Step Links
- wikiHow categories hierarchy
A multilingual version of this data can be found on Kaggle. This multilingual dataset contains over 800K instructions in 16 different languages.
The HowLinks tool can be used to visualise this dataset and its links to DBpedia in a web browser.
A live demo of this tool (although only serving a subset of this dataset) is available here.
- 23,033,490: number of triples.
- 2,610,223: number of labelled RDF nodes.
- 215,959: number of instructions. 77% from wikiHow (datasets 1of11_knowhow_wikihow to 9of11_knowhow_wikihow) and 23% from Snapguide (datasets 10of11_knowhow_snapguide to 11of11_knowhow_snapguide).
- 255,101: number of process inputs linked to 8,453 distinct DBpedia concepts (dataset Process - Inputs)
- 4,467: number of process outputs linked to 3,439 distinct DBpedia concepts (dataset Process - Outputs)
- 193,701: number of step links between 114,166 different sets of instructions (dataset Process - Step Links)