Welcome to the documentation of the EUROSENTIMENT project¶

The EUROSENTIMENT format for services and corpora¶

The Eurosentiment format is an extension of the NIF format data model for use in Sentiment Analysis. However, NIF and the Eurosentiment differ in one respect: Eurosentiment sets JSON-LD as its primary serialisation format, whereas NIF defaults to XML+RDF or turtle. It includes properties from Marl, Onyx and other ontologies that complement those in NIF for sentiment and emotion tagging. However, NIF and the Eurosentiment differ in one respect: Eurosentiment sets JSON-LD as its primary serialisation format, whereas NIF defaults to XML+RDF or turtle.

JSON-LD is a subset of JSON that makes it possible to embed semantic information in plain JSON objects. It retains full compatibility with JSON while adding useful information.

By using this serialisation format, Eurosentiment targets both semantic web developers and traditional developers alike.

{
  "@context": [
    "http://demos.gsi.dit.upm.es/eurosentiment/static/context.jsonld",
],
"@id": {{ processID }},
"analysis": [
  {
    "@id": {{ analysisID }},
    "@type": [
      {{ analysisType }}
    ],
    "prov:wasAssociatedWith": {{ agent }},
    "dc:language": {{ language}},
    "marl:maxPolarityValue": {{ minValue }},
    "marl:minPolarityValue": {{ maxValue }}
  }
  [...]
],
"domain": {{ domain }},
"entries": [
  {
    "@id": {{ entry_id }},
    "dc:subject": {{ topic }},
    "emotions": [
      {
        "prov:generatedBy": {{ analysisID }},
        "onyx:hasEmotion": [
          {
            "onyx:hasEmotionCategory": {{ emotions[i].category }},
            "onyx:intensity": {{ emotions[i].emotion_intensity }}
          },
          [...]
        ]

      }
      [...]
    ],
    "opinions": [
      {
        "prov:generatedBy": {{ analysisID }},
        "marl:polarityValue": {{ opinions[i].polarityValue }},
        "marl:hasPolarity": {{ opinions[i].polarity }},
        "marl:describesObject": {{ opinions[i].described_object }},
      },
      [...]
    ],
    "nif:isString": {{ string_representation }},
    "strings": [
      {
        "nif:anchorOf": {{ strings[i].value }},
        "itsrdf:taIdentRef": {{ strings[i].entity }},
        "nif:posTag": {{ strings[i].posTag }},
        "nif:lemma": {{ strings[i].lemma }}
      },
      [...]
    ]
  },
  [...]
]
}

{
  "@context": {
      "dc": "http://purl.org/dc/terms/",
      "dc:subject": {
        "@type": "@id"
      },
      "emotions": {
        "@container": "@list",
        "@id": "onyx:hasEmotionSet",
        "@type": "onyx:EmotionSet"
      },
      "marl": "http://www.gsi.dit.upm.es/ontologies/marl#",
      "nif": "http://persistence.uni-leipzig.org/nlp2rdf/ontologies/nif-core#",
      "onyx": "http://www.gsi.dit.upm.es/ontologies/onyx#",
      "opinions": {
        "@container": "@list",
        "@id": "marl:hasOpinion",
        "@type": "marl:Opinion"
      },
      "prov": "http://www.w3.org/ns/prov#",
      "rdfs": "http://www.w3.org/2000/01/rdf-schema#",
      "analysis": {
        "@id": "prov:wasInformedBy"
      },
      "entries": {
        "@id": "prov:generated"
      },
      "strings": {
        "@reverse": "nif:hasContext",
        "@type": "nif:String"
      },
      "wnaffect": "http://www.gsi.dit.upm.es/ontologies/wnaffect#",
      "xsd": "http://www.w3.org/2001/XMLSchema#"
  }
}

{
 "@context": [
   "http://demos.gsi.dit.upm.es/eurosentiment/static/context.jsonld"
 ],
 "results": {
   "analysis": [
     {
       "@id": "http://example.com/analyse",
       "@type": [
         "marl:SentimentAnalysis"
       ],
       "dc:language": "en",
       "marl:maxPolarityValue": 10.0,
       "marl:minPolarityValue": 0.0
       "prov:wasAssociatedWith": "http://example.com"
     }
   ],
   "entries": [
     {
       "@id": "http://example.com/analyse?input=My%20ipad%20is%20an%20awesome%20device",
       "opinions": [
         {
           "marl:polarityValue": 9,
           "marl:hasPolarity": "marl:Positive",
           "marl:describesObject": "http://dbpedia.org/page/IPad"
           "prov:generatedBy": "http://example.com/analyse",
         }
       ],
       "nif:isString": "My ipad is an awesome device",
       "strings": [
         {
           "@id": "http://example.com/analyse?input=My%20ipad%20is%20an%20awesome%20device#char=3,6",
           "nif:anchorOf": "ipad",
           "itsrdf:taIdentRef": "http://dbpedia.org/page/IPad"
         }
       ]
     }
   ]
 }
}

{
 "@context": [
   "http://demos.gsi.dit.upm.es/eurosentiment/static/context.jsonld"
 ],
 "results": {
   "analysis": [
     {
       "@id": "http://example.com/analyse",
       "@type": [
         "onyx:EmotionAnalysis"
       ],
       "dc:language": "es",
       "onyx:maxEmotionIntensity": 1.0,
       "onyx:minEmotionIntensity": 0.0
       "prov:wasAssociatedWith": "http://example.com/"
     }
   ],
   "entries": [
     {
       "@id": "http://example.com/analyse?input=Mi%20ipad%20me%20tiene%20harto",
       "dc:language": "es",
       "opinions": [
       ],
       "emotions": [
         {
           "onyx:aboutObject": "http://dbpedia.org/page/IPad"
           "prov:generatedBy": "http://example.com/analyse",
           "onyx:hasEmotion": [
             {
                 "onyx:hasEmotionCategory": "wna:dislike",
                 "onyx:hasEmotionIntensity": 0.7
             },
             {
                 "onyx:hasEmotionCategory": "wna:despair",
                 "onyx:hasEmotionIntensity": 0.1
             }
           ]
         }
       ],
       "nif:isString": "My ipad is an awesome device",
       "prov:generatedBy": "http://example.com/analyse",
       "strings": [
         {
           "@id": "http://example.com/analyse?input=Mi%20ipad%20me%20tiene%20harto#char=3,6",
           "nif:anchorOf": "ipad",
           "itsrdf:taIdentRef": "http://dbpedia.org/page/IPad"
         }
       ]
     }
   ]
 }
}

The Eurosentiment Format for Lexica¶

The EUROSENTIMENT sentiment lexicons are represented in RDF using the lemon format extended with some properties from the Marl vocabulary for representing the sentiment information such as polarity and polarity value. Deliverable D4.3 (p54-56) describes in details the model.

The following figure shows an example of RDF domain-specific lexicon.

In this figure we see snippets from two lexicons: a lexicon for the hotel domain in english (i.e. le:hotel ) and a lexicon for the hotel domain in german (i.e. ld:hotel). The sentiment lexicons are composed of lexical entries: in our example lee:location and lee:pretty-good for the English lexicon and led:Lage for the german lexicon. Some of the lexical entries are domain aspects like location and Lage and some are sentiment words like pretty good.

Each lexical entry is defined by a lemon canonicalForm, several lemon otherForm properties representing the morphological variations, several senses and part of speech information. The connection of the EUROSENITMENT lexicons to other linguistic linked data datasets (i.e. DBpedia and Wordnet) happen at the sense level. For instance in our case, the sens of the lee:location lexical entry is linked to the http://dbpedia.org/page/Location entity and to the Wordnet synset: http://wordnet-rdf.princeton.edu/wn31/100027365-n.

In the case of sentiment lexical entries we use two more properties (i.e. marl:PolarityValue and marl:hasPolarity) to represent the sentiment inforation. In our example, lee:pretty-good has a positive polarity of 0.75 where the most positive value is 1 and the most negative value is -1.

The domain-specific aspect of the EUROSENTIMENT lexicons is given by the lemon property context which connectsa sentiment word to a domain aspect. In the example, the lmn:context property signifies that the sentiment word pretty good has a positive value of 0.75 in the context of the location domain aspect.

We recall from D4.3 that we generate domain-specific lexicons in other languages based on the initial english lexicon. For example, in our case, the german lexicon for hotel was automatically generated form the english lexicon for hotel. This relation between the two lexicons is represneted via the isocat:translationOf relation between the senses of the translated lexical entries.

Eurosentiment Corpus Converter¶

Architecture¶

As can be seen in the figure below, the Corpus Converter is made out of A Web Server proxies all the user requests to the Request Processor. This processor is a Django application that has two roles: providing a REST and an Administrative interface. On the administrative side it deals with authorisation and authentication of users. It also stores the translation templates provided by the staff in the database On the REST side, it forwards all valid user translation requests to the Translator module. This module opens the original file in any of the supported formats, it applies the requested translation template with its content, and then returns the resulting document.

Translating a document¶

Documents can be translated via the Web Interface or using the REST interface. Actually, the form in the Web is simply a convenient way of accessing the REST interface which shows all the available templates and a field to upload the desired file.

Translating a document through the web

The Corpus Converter endpoint takes the following parameters:

• input (i): The original file to be translated
• informat (f): The format of the original file
• intype (t) [Optional]:
  • direct (default)
  • url
  • file
• outformat (o):
  • json-ld
  • rdfxml
  • turtle (default, to comply with NIF)
  • ntriples
  • trix
• base URI (u) [Optional]: base URI to use for the corpus
• prefix (p) [Optional]: prefix to replace the base URI
• template (t) [Optional]: ID of the template to use. If it is omitted, a template to convert from informat to outformat will be used, or a template from informat to another format (e.g. json-ld), with automatic conversion.
• toFile [Optional]: Whether the result should be sent in the response (default) or written to a file. For convenience, this value defaults to False when using the Web Form.

Using the command line tool curl, a request can be made like this:

curl http://demos.gsi.dit.upm.es/eurosentiment/marlgenerator/process -F"intype=file" -F"informat=Example" -F"outformat=jsonld" -F"input=@input-file.csv" > result.jsonld

Adding a template¶

Editing a template is simple. First, visit the administration URL. If it is your first login or if your session expired, you will be greeted by a login screen:

Login prompt

Just enter your username and password, and the administration interface should appear.

Administration Interface

Editing a template

It is also possible to add a format from this menu, clicking on the “Plus” icon:

Adding a format on the fly

Checking usage statistics¶

Once logged in as a superuser, you can also add new users and check the requests that have been made for each format.

Superuser panel

To check the requests, click on “Translation Requests” in the administration panel.

Log of requests

In addition to simply checking the requests, it is also possible to filter the requests using different parameters. This feature is especially useful if you want to study the popularity of a format, or to compare different templates for the same formats.

Filtering requests

{
    "@context": [
        "http://demos.gsi.dit.upm.es/eurosentiment/static/context.jsonld",
    ],
    "@id": "{{ linesplit(f.name,"/")[-1] }}",
    "analysis": [
        {
            "@id": "{{ linesplit(f.name,"/")[-1] }}#MachineAnnotated",
            "@type": [
                "marl:SentimentAnalysis"
            ],
        {% if language %}
            "dc:language": "{{ language}}",
        {% endif %}
            "marl:maxPolarityValue": 10.0,
            "marl:minPolarityValue": 0.0,
            "prov:wasAssociatedWith": "pt:agent"
        }
    ],
    "entries": [
{% for line in f %}
{% set i=linesplit(line, "\t") %}
{% set node="_:BlankNode%s" % loop.index %}
{% set text = i[0] %}
{% set syntax=linesplit(i[1][1:-1], ",") %}
{% set pol= i[2] | float %}
        {
            "@id": "{{ node }}",
            "opinions": [
                {


{% if pol%}
                     "marl:polarityValue": {{ pol }},
{% if pol > 5 %}
                     "marl:hasPolarity": "marl:Positive"
{% elif pol < 5 %}
                     "marl:hasPolarity": "marl:Negative"
{% else %}
                     "marl:hasPolarity": "marl:Neutral"
{% endif %}
{% endif %}
                }
            ],
            "nif:isString": {{ text | escapejs }},
            "strings": [
              {% for s in syntax %}
              {
              {% set parts=linesplit(s, ";;") %}
              "nif:anchorOf": {{ parts[0] | escapejs }},
              "nif:posTag": "pt:{{ parts[1] }}",
              "nif:lemma": {{ parts[2] | escapejs }} }
              {% if not loop.last %}, {% endif %}{% endfor %}
            ]
        } {% if not loop.last %} , {% endif %}
{% endfor%}
    ]
}

Eurosentiment Playground¶

EUROSENTIMENT provides services and resources for Sentiment Analysis in several languages. There are several utilities, code snippets and instructions on how to make use of the platform publicly available. However, all of them require the installation of a third party tool or the use of a programming language to consume the API. The EUROSENTIMENT Playground solves this problem by providing an easy-to-use web interface to make API calls. Read our simple instructions and start using EUROSENTIMENT today!

The playground is available here: http://demos.gsi.dit.upm.es/eurosentiment-playground/

Language Resource Adaptation Pipeline¶

In this section we describe the Language Resource Adaptation Pipeline various components and provides links to the source code for these components.

The Language Resource Adaptation Pipeline (a.ka. LRAP) implements a methodology for legacy language resource adaptation that generates domain-specific sentiment lexicons organized around domain entities described with lexical information and sentiment words described in the context of these entities.

The outcome of the Language Resource Adaptation Pipeline are annotated corpora represented the NIF/Marl format and domain-specific sentiment lexicons represented in RDF using the Lemon/Marl format. The legacy language resources are enriched with semantics and additional linguistic information from resources like DBpedia and BabelNet.

There are four main steps of the LRAP as shown in the Figure bellow:

1. Corpus Conversion: normalizes the different language resources to a common schema based on Marl and NIF; The corpus convertor tool was described earlier in a separate section.
2. Semantic Analysis: extracts the domain-specific entity classes and named entities and identifies links between these entities and concepts from the LLOD Cloud. The Semantic Analysis step consists of: Domain Modeller (DM), Entity Extraction (EE), Entity Linking (EL) and Synset Identification (SI) components.
3. Sentiment Analysis: extracts contextual sentiments and identifies SentiWordNet synsets corresponding to these contextual sentiment words. The Sentiment Analysis step consists of: Domain-Specific Sentiment Polarity Analysis (DSSA) and Sentiment Synset Identification (SSI) components.
4. Lexicon Generator: uses the results of the previous steps, enhances them with multilingual and morphosyntactic information and converts the results into a lexicon based on the lemon and Marl formats. The Lexicon Generator step consists of: MorphoSyntactic Enrichment (ME), Machine Translation(T) and lemon/Marl Generator(LG) components.

Different language resources are processed with variations of the given adaptation pipeline. For more details on the LRAP and the domain-specific lexicons it generates please check our dissemination material:

• Presentatin at 5th International Workshop on EMOTION, SOCIAL SIGNALS, SENTIMENT & LINKED OPEN DATA: “Generating Linked-Data based Domain-Specific Sentiment Lexicons from Legacy Language and Semantic Resources” - Gabriela Vulcu, Paul Buitelaar, Sapna Negi, Bianca Pereira, Mihael Arcan, Barry Coughlan, Fernando J. Sanchez and Carlos A. Iglesias
• Poster at the Data Challenge at the 3rd Workshop on Linked Data in Linguistics “Linked-Data based Domain-Specific Sentiment Lexicons” - Gabriela Vulcu, Raul Lario Monje, Mario Munoz, Paul Buitelaar and Carlos A. Iglesias

How to add a new service¶

This section describes how a new service is added to the EUROSENTIMENT LRP.

Service upload steps:

0. Precondition: the user is logged in: https://portal.eurosentiment.eu/ with his username and password

1. Click on the ‘Services’ tab -> ‘Add a service’

2. Fill in the ‘Service creation form’

   • Name -> give a name of your service
   • Description -> add a detailed description of your service
   • HTTP method -> select the HTTP method and fill in the service’s access URI
   • Credentials -> if the serviec endpoit is not public, please provide an access token
   • Request cost -> fill in the cost per request to your service
   • Request limit -> fill in teh limit of requests a user of the service should not exceed
   • Language -> select form the language drop-down list
   • Domain -> select form the drop-down list
   • Fill in the Fact sheet with similar details and send an email to eurosentimentpt@gmail.com

3. Click on the ‘Create’ button

NIF API¶

GET /services/server/access/(service_id)¶

Access the service at service_url. The service_id can be retrieved from the service page in the Portal Since the requests to the server are likely to be long, POST /services/server/access/(service_id) is recommended.

Example request:

GET /service/access/SentimentAnalysisExample?input=I%20love%20EUROSENTIMENT HTTP/1.1
Host: eurosentiment.eu
Accept: application/json, text/javascript

Example response:

HTTP/1.1 200 OK
Vary: Accept
Content-Type: text/javascript

{
  "@context": [
    "http://eurosentiment.eu/context.jsonld",
    {
      "@base": "http://eurosentiment.eu/service/access/SentimentAnalysisExample#"
    }
],
"analysis": [
  {
    "@id": "SentimentAnalysisExample",
    "@type": "marl:SentimentAnalysis",
    "dc:language": "en",
    "marl:maxPolarityValue": 10.0,
    "marl:minPolarityValue": 0.0
  }
],
"domain": "wndomains:electronics",
"entries": [
  {
    "opinions": [
      {
        "prov:generatedBy": "SentimentAnalysisExample",
        "marl:polarityValue": 7.8,
        "marl:hasPolarity": "marl:Positive",
        "marl:describesObject": "http://eurosentiment.eu",
      }
    ],
    "nif:isString": "I love EUROSENTIMENT",
    "strings": [
      {
        "nif:anchorOf": "EUROSENTIMENT",
        "nif:taIdentRef": "http://eurosentiment.eu"
      }
    ]
  }
 ]
}

Query Parameters:
	input (i) – No default. Depends on informat and intype informat (f) – one of turtle (default), text, json-ld intype (t) – one of direct (default), url outformat (o) – one of turtle (default), text, json-ld prefix (p) – prefix for the URIs
Request Headers:
	Accept – the response content type depends on Accept header X-Eurosentiment-Token – optional OAuth token to authenticate
Response Headers:
	Content-Type – this depends on Accept header of request
Status Codes:	200 OK – no error 404 Not Found – service not found

POST /services/server/access/(service_id)¶

The same as the previous method. This is the recommended method.

Form Parameters:
	i/input – No default. Depends on informat and intype f/informat – one of turtle (default), text, json-ld t/intype – one of direct (default), url o/outformat – one of turtle (default), text, json-ld p/prefix – prefix for the URIs
Request Headers:
	Accept – the response content type depends on Accept header X-Eurosentiment-Token – optional OAuth token to authenticate
Response Headers:
	Content-Type – this depends on Accept header of request
Status Codes:	200 OK – no error 404 Not Found – service not found

Create your own services¶

If, instead of using any of the provided services you want to roll your own, you can also contribute to the Eurosentiment pool by publishing your service. The Eurosentiment github repository contains a series of tutorials that will help you get started with the process, including some complete examples of sentiment analysis service in different programming languages.

As of this writing, there are examples in:

• Java: https://github.com/eurosentiment/eurosentiment-tutorial-java
• Python: https://github.com/eurosentiment/eurosentiment-tutorial-python
• Ruby: https://github.com/eurosentiment/eurosentiment-tutorial-ruby

How to add a new language resource¶

This section describes how a new language resource is added to the EUROSENTIMENT LRP. You can also watch a video (https://www.dropbox.com/s/86dqo4u4k9gf0i6/EUROSENTIMENT-adding-a-new-LR.mp4?dl=0) that describes step-by-step the process.

Resource upload steps:

0. Precondition: the user is logged in: https://portal.eurosentiment.eu/ with his username and password

1. Click on the ‘Recources’ tab -> ‘Add a resource’

2. Fill in the ‘Resource creation form’

   • Name -> give a name of your LR
   • Graph URI prefix -> the platform will suggest you a unique graph URI (if you like it, leave it like it is)
   • Short Name -> give a short name of your LR. It will be used to dynamically generate the graph URI
   • Language -> select form the language drop-down list
   • Application Domain -> select form the drop-down list
   • Description -> add a detailed description of your language resource.
   • Resource type -> chose from the drop-down list the resource type you want to add.
   • Access control -> select the type of license under which you want to publish the LR
   • Upload the LR file
   • Fill in the Fact sheet with similar details and send an email to eurosentimentpt@gmail.com

3. Click on the ‘Create’ button

4. What will happen behind the scenes:

   • an email is sent to the EUROSENTIMENT LRP
   • you are notified by email that your LR was submitted and will be reviewed before added to the LRP as soon as possible
   • an administrator will carefully read the resource submission and will decide if the language resource will be added to the language resource pool
   • if the LR is provided in an existing format then move to 4.6
   • if the LR is provided in a new format not known previously to the platform, the administrator will develop a specific language resource adaptation pipeline
   • the administrator runs the language resource adaptation pipeline and the provided LR is processed, converted to RDF and linked-data and the result is uploaded to the SPARQL endpoint
   • you are notified by email that your language resource was processed

5. Click on the ‘Resources’ -> ‘List Own resources’ and see your newly added LR that can be used by you or other users of the EUROSENTIMENT LRP

Resources API¶

POST /sparql¶

The posts tagged with tag that the user (user_id) wrote.

Example request:

POST /sparql HTTP/1.1
Host: eurosentiment.eu
Content-Length: 199
x-eurosentiment-token: 23aee871-d18d-4afa-c2e3-283f8ae9232ca
Accept-Encoding: gzip, deflate, compress
Accept: */*
content-type: application/json

{"query": "PREFIX lemon: <http://lemon-model.net/lemon#>\nSELECT * FROM <http://www.eurosentiment.eu/dataset/hotel/es/gabvul/9/lexicon>\nWHERE {?s lemon:sense ?sense }", "format": "application/json"}

Example response:

HTTP/1.1 200 OK
date: Mon, 07 Jul 2014 16:46:21 GMT
content-length: 596
content-type: application/json
server: Jetty(8.1.10.v20130312)

{
    "head": {
        "link": [],
        "vars": [
            "s",
            "sense"
        ]
    },
    "results": {
        "bindings": [
            {
                "s": {
                    "type": "uri",
                    "value": "http://www.eurosentiment.eu/dataset/hotel/es/gabvul/9/lexicalentry/H2O"
                },
                "sense": {
                    "type": "uri",
                    "value": "http://www.eurosentiment.eu/dataset/hotel/es/gabvul/9/lexicalentry/sense/H2O_0"
                }
            },
            {
                "s": {
                    "type": "uri",
                    "value": "http://www.eurosentiment.eu/dataset/hotel/es/gabvul/9/lexicalentry/a_gusto"
                },
                "sense": {
                    "type": "uri",
                    "value": "http://www.eurosentiment.eu/dataset/hotel/es/gabvul/9/lexicalentry/sense/a_gusto_0"
                }
            }
        ],
        "distinct": false,
        "ordered": true
    }
}

:data :reqheader Accept: the response content type depends on

Accept header

Request Headers:
	Authorization – optional OAuth token to authenticate
Response Headers:
	Content-Type – this depends on Accept header of request
Status Codes:	200 OK – no error 404 Not Found – there’s no user

Language Resource Types¶

This section describes the type of language resources that the EUROSENTIMENT LRP supports. Based on the initial list of language resources (see here) from the project we identified the following supported formats. please not in brackets the number of the Lnagua resource addressed by the listed format.

1. domain-specific-lexicon-TSV (5): Tab-separated-values file that describes sentiments in the context of domain aspects (e.g. myFile.tsv). The header of the TSV file should have the following columns:

entityWNid    entityPOS    entity    sentiWNid    sentiPOS    sentiment    sentiScore

where:

• entityWNid: WordNet30 synset ID of the domain aspect (e.g. 02671062).
• entityPOS: part of speech of the domain aspect (i.e. n, a, v, r).
• entity: domain aspect as string (e.g. “access”).
• sentiWNid: WordNet30 synset ID of the sentiment associated with the domain aspect (00979366).
• sentiPOS: part of speech of the sentiment word (i.e. n, a, v, r).
• sentiment: sentiment word as string (e.g. “quick”)
• sentiScore: the polarity value of the sentiment (a rational number between -1 and 1, where -1 is very negative, 0 is neutral and 1 is very positive).

e.g.

02671062    n    access    00979366    a    quick    0.5

2. entiment-lexicon-CSV (45 (needs to be converted), 54): Comma-separated-values file that describes the sentiment words and their polarities from a domain. (e.g. myFile.csv). The header of the CSV file should have the following columns:

sentiment,sentimentPOS,sentimentScore,morphosyntacticVariations

where:

• sentiment: the sentiment word in the domain.
• sentimentPOS: the part-of-speech of the sentiment word.
• sentimentScore: the polarity value of the sentiment (a rational number between -1 and 1, where -1 is very negative, 0 is neutral and 1 is very positive).
• morphosyntacticVariations:the sentiment morphosyntactic variations.

e.g.

Besserung|NN    0.40    Besserungen,Besserungnen

3. review-corpus-no-polarity (36): A file containing one review per line.

e.g.

The location was great and the staff friendly. I like it!

The room was a bit too small.

…

4. review-corpus-overall-polarity-TSV (3, 31, 34): A tab-separated-values file with reviews and overall polarity. The header of the TSV file should have the following columns:

reviewText    overallPolarity

where:

• reviewText: A string that contains the review (no tabs in the string).
• overallPolarity: The overall polarity of the given review text (rational number between -1 and 1, where -1 is very negative, 0 is neutral and 1 is very positive).

e.g.

Rien à dire. Très bon produit de qualité.    1.0

5. review-corpus-pos-lemma-wn-overall-polarity-TSV (2, 4): A tab-separated-values file with columns of the following form:

reviewText    annotation    overallPolarity

where:

• reviewText: A string that contains the review (no tabs in the string).
• annotation: A list with comma-separated values for each word in the text review containing: word, part-of-speech, lemma and a list of possible WordNet30 synsteIDs.
• overallPolarity: The overall polarity of the given review text (rational number between -1 and 1, where -1 is very negative, 0 is neutral and 1 is very positive).

e.g.

Excellent location.    [Excellent;;JJ;;excellent;;[], location;;NN;;location;;[n#01051331]]    0.8

6. review-corpus-pos-lemma-wn-overall-polarity-Excel (7,8,9,10,11,12,13,14,17,18,20,21,22,50,51,52,53): A .gz compressed folder containing Excel files with several reviews per file. Each review in the Excel file is spread over several lines. The header of the Excel file is: TEXT LEMMA WN_POS WN_SYNSET DOMAIN SENTIMENT EMOTION.

where:

• TEXT: the full review text in the first line; subsequent lines have one word of the review per line.
• LEMMA: nothing in the first line; in subsequent lines it describes the lemma value of an individual word from the review.
• WN_POS: nothing in the first line; in subsequent lines it describes the part-of-speech value of an individual word from the review.
• WN_SYNSET: nothing in the first line; in subsequent lines it describes the WordNet30 synset ID value of an individual word from the review.
• DOMAIN: the domain of the review (only in the first line)
• SENTIMENT: (only in the first line)
• EMOTION: (only in the first line)

7. dataset-rdf (25, 26, 27, 29, 37(needs conversion to RDF),42,55): RDF dump (*.nt.gz) of linked data dataset like WordNet, DBpedia, BabelNet.
8. aspects-review-corpus-TripAdvisor (49): A file with annotated reviews and aspect ratings. Each review in the file is spread over several lines where each line starts with a dedicated tag as in the example below.

e.g.

<Author>IndieLady

<Content>Lovely hotel, unique decor, friendly front desk staff […]

<Date>Nov 13, 2008

<No. Reader>-1

<No. Helpful>-1

<Overall>4

<Value>5

<Rooms>4

<Location>5

<Cleanliness>4

<Check in / front desk>5

<Service>5

<Business service>-1

9. aspects-review-corpus-Amazon (44): A file that consists of plain text reviews for products with custom ratings annotations that spread over several lines. The marker for a new review is [t] whereas the numbers in brackets stand for the rating of a certain aspect in the review. See below an example:

e.g.

[ t ] the best 4mp compact digital available camera[+2]## this camera is perfect for an enthusiastic amateur photographer . picture [+3] ,

macro[+3]## the pictures are razor sharp , even in macro . . .

10. Opener lexicon : Semicolon-separated-values file with the following columns:

wordnetSynsetID; POStag; polarity; confidence; lemmas; manualReviewFlag

where:

• wordnetSynsetID : wordnet 30 synsetID.
• POStag : part-of-speech tag
• polarity : sentiment polarity which can be -1, 0 or 1 for negative, neutral and positive respectively.
• confidence : confidence assigned by the propagation algorithm
• lemmas : lemmas of this synset in wordnet separated by comma
• manualReviewFlag : -1 if no manual review has been done and + 1 if they have been reviewed.

e.g.

eng-30-09366317-n;n;positive;0.3125;natural_elevation,elevation;-1

eng-30-07961016-n;n;neutral;0.3125;clod,glob,ball,chunk,clump,lump;-1