We annotate the Rest14 and Rest15 datasets. 

The train, dev, and test sets in each dataset include single-aspect and multi-aspect sentences, respectively.

We annotate multi-aspect sentences with overlapping (1) or non-overlapping (0). 
 
1. In Rest14, the dataset is in the format of XML. An example is given below. 
Two aspect categories "food" and "service" are non-overlapping. 
Note that we only use the aspect category information in our experiments.

############################################################################
<sentence id="41">
    <text>horrible food and horrible service.</text>
    <aspectTerms>
        <aspectTerm term="food" polarity="negative" from="9" to="13"/>
        <aspectTerm term="service" polarity="negative" from="27" to="34"/>
    </aspectTerms>
    <aspectCategories>
        <aspectCategory category="food" polarity="negative"/>
        <aspectCategory category="service" polarity="negative"/>
    </aspectCategories>
    <aspectCategoriesOverlap>0</aspectCategoriesOverlap>
</sentence>
############################################################################

2. In Rest15, the dataset is in the format of plain text. An example is given below. 
The first line includes the information of sentence, aspect categories, and their sentiment polarities. They are separated by ####.
The second line is for overlapping/non-overlapping annotation. 
In this example, the aspect categories "DRINKS#QUALITY" and "SERVICE#GENERAL" are overlapping.

############################################################################
The wine the service was very good too.####DRINKS#QUALITY SERVICE#GENERAL ####positive positive
aspectCategoryOverlap----1
############################################################################