Participants who want to submit their systems need to follow the steps below:

1. Send an email to eplopes@inf.ufpel.edu.br or asqppt@inf.ufpel.edu.br with the team info (email and affiliation of each team member)

2. Organizers will send you the training data by February 17th. 

3. Organizers will send you the test data by March 10th.

4. Participants send their predictions by May 5th and an draft of your working notes (Submission form LINK).

5. Once the results are declared (May 5th), participants will be required to provide an abstract and a technical report including a brief description of their approach and experiments for the publication in the IberLEF Proceedings. 

The results should be sent in the following format:

Task 1 - Aspect Term Extraction (ATE): The input will be only the review, and the expected output is a list of the aspect terms present on the review. 

For example, given the input “O hotel é caro, mas os quartos são grandes e limpos” (“the hotel is expensive, but the rooms are big and clean”), the expected output would be [“hotel”, “quartos”] ([“hotel”, “rooms”])

Task 2 - Opinion Term Extraction (OTE): The input will be the review and a single aspect term, and the expected output is a list of sentiment terms related to the aspect. 

Using the same example from the previous task, for the aspect “hotel”, the expected output would be just [“caro”] ([“expensive”]); and for the aspect “quartos”, the expected output would be [“grandes”, “limpos”] ([“big”, “clean”])

Task 3 - Aspect Category Detection (ACD): The input will be the review and a single aspect, and the expected output is the label of the predicted category (in total, six classes are possible). 

Using the same example, the expected outputs of this task would be “price” and “structure”, for the aspects “hotel” and “rooms” respectively. These are two of the six possible categories.

Task 4 - Aspect Sentiment Quad Prediction (ASQP): For this task, the input is a single review text, and the expected output is a list of quadruples. Each quadruple is described above. Still using the same input, the expected output would be a list containing the following quadruplets:

{"category": "price", "aspect": "hotel", “sentiment": "caro", "polarity": "NEG"}

{"category": "structure", "aspect": "quartos", “sentiment": "grandes", "polarity": "POS"}

{"category": "structure", "aspect": "quartos", “sentiment": "limpos", "polarity": "POS"}