-
Notifications
You must be signed in to change notification settings - Fork 2
Usage Guide
Here are listed some definitions useful to understand the following paragraphs.
- Seed: Twitter handler representing the starting point of the process. It acts as samples of concepts to search for.
- Expert Type: DBPedia type. It's used as a descriptor of the domain of interest.
- Candidate: it's an entity retrieved by the process. It's associated with a numerical score indicating how much is similar to the starting seeds.
SKE comes with three pre-configured scenarios you can run. They have the purpose to showcase the capabilities of the system.
This scenario involve entities from the Fashion domain. In particular:
- Seeds: Twitter handle belonging to emerging fashion designer
- Expert types: "Artist", "Broadcaster", "Magazine", "Model", "Organisation" and "TelevisionShow"
This scenario involves writers engaged in the Melbourne Emerging Writers Festival, with the objective of discover other emerging writers.
- Seeds: Twitter handle belonging to writers.
- Expert types: "Organisation","WrittenWork","PeriodicalLiterature","Magazine","Website","Writer","Book","Newspaper","Politician","Event","Philosopher"
This scenario involves journalist and bloggers expert of the finance sector. The object is to find other other influencers in finance.
- Seeds: Twitter handle belonging to blogger and journalists in the finance sector.
- Expert types: "Economist", "Journalist", "Organisation", "Bank", "Politician"
In order to build an experiment, you need to provide these input elements:
- A title for the experiment.
- The list of seeds, i.e., the twitter user ids (handles without the @) that represent the emerging entities for your field of interest (up to 20).
- The list of expert types that are relevant for your domain (you can search in the list or directly check them in the checklist).
Once launched, the pipeline go through three states:
- PROCESSING: the pipeline was created and it's waiting to be executed.
- CRAWLING: SKE is crawling Twitter and calling Dandelion to retrieve the mentions and the entity
- COMPUTING_CANDIDATES: SKE is using the mentions to compute the feature vectors and calculate the similarity in order to rank the candidates.
- COMPLETED: the pipeline is completed and the results can be seen.
Once the pipeline is completed, the results (e.g., the list of extracted candidates) can be seen in the experiment details page.
Initially the page shows the top 20 candidates sorted according to their similarity score (i.e., 1 same behavior as the seeds, 0 totally different behavior). By clicking More results, it will load the next 20 candidates.
By clicking the Export button you can download the full list of candidates.
SKE allows to forward the results to domain experts or a generic crowd for evaluation purposes. By clicking on Evaluation Page button you will be brought to an evaluation page that you can share to make people verify the correctness of the results.
The outcome of the evaluation can be see in the experiment details page.
The extracted candidates can be used as new seeds to run the whole pipeline from the beginning.
To do so select the candidates by checking the Set as new Seed checkbox and then clicking the Run with new seeds button. The pipeline will run again with the same expert types and the new selected seeds.