User feedback analysis in KBC using Geneea NLP platform

Integration of the Geneea API with Keboola Connection.

This is a Docker container used for running user-feedback NLP analysis jobs in the KBC. Automatically built Docker images are available at Docker Hub Registry.

Building a container

To build this container manually one can use:

git clone https://github.com/Geneea/kbc-feedback-analysis.git
cd kbc-feedback-analysis
sudo docker build --no-cache -t geneea/kbc-feedback-analysis .

Running a container

This container can be run from the Registry using:

sudo docker run \
--volume=/home/ec2-user/data:/data \
--rm \
geneea/kbc-feedback-analysis:latest

Note: --volume needs to be adjusted accordingly.

Sample configuration

Mapped to /data/config.json

{
  "storage": {
    "input": {
      "tables": [
        {
          "destination": "comments.csv"
        }
      ]
    }
  },
  "parameters": {
    "user_key": "<ENTER API KEY HERE>",
    "columns": {
      "id": ["feedback_id"],
      "text": ["summary"],
      "positives": ["pos_1", "pos_2"],
      "negatives": ["neg_1", "neg_2"]
    },
    "language": "cs",
    "domain": "retail",
    "feedback_entities": ["service", "product"],
    "feedback_relations": ["ATTR"],
    "correction": "aggresive",
    "diacritization": "auto",
    "use_beta": false
  }
}

Output format

The results of the NLP analysis are written into five tables.

• analysis-result-comments.csv with comment-level results in the following columns:

  • all id columns from the input table (used as primary keys)
  • language detected language of the comment, as ISO 639-1 language code
  • sentimentValue detected sentiment of the comment, from an interval [-1.0; 1.0]
  • sentimentPolarity detected sentiment of the comment (-1, 0 or 1)
  • sentimentLabel sentiment of the comment as a label (negative, neutral or positive)
  • usedChars the number of characters used by this comment

• analysis-result-sentences.csv with sentence-level results has the following columns:

  • all id columns from the input table (used as primary keys)
  • index zero-based index of the sentence in the comment, (primary key)
  • segment text segment where the sentence is located
  • text the sentence text
  • sentimentValue detected sentiment of the sentence, from an interval [-1.0; 1.0]
  • sentimentPolarity detected sentiment of the sentence (-1, 0 or 1)
  • sentimentLabel sentiment of the sentence as a label (negative, neutral or positive)

• analysis-result-entities.csv with entity-level results has the following columns:

  • all id columns from the input table (used as primary keys)
  • type type of the found entity, e.g. person, organization or tag, (primary key)
  • text disambiguated and standardized form of the entity, e.g. John Smith, Keboola, safe carseat, (primary key)
  • score relevance score of the entity, e.g. 0.8
  • entityUid unique ID of the entity, may be empty
  • sentimentValue detected sentiment of the entity, from an interval [-1.0; 1.0]
  • sentimentPolarity detected sentiment of the entity (-1, 0 or 1)
  • sentimentLabel sentiment of the entity as a label (negative, neutral or positive)

• analysis-result-relations.csv with relations-level results has the following columns:

  • all id columns from the input table (used as primary keys)
  • type type of the found relation, VERB or ATTR, (primary key)
  • name textual name of the relation, e.g. buy or smart, (primary key)
  • subject possible subject of the relation (primary key)
  • object possible object of the relation (primary key)
  • subjectType type of the relation's subject
  • objectType type of the relation's object
  • subjectUid unique ID of the relation's subject
  • objectUid unique ID of the relation's object
  • sentimentValue detected sentiment of the relation, from an interval [-1.0; 1.0]
  • sentimentPolarity detected sentiment of the relation (-1, 0 or 1)
  • sentimentLabel sentiment of the relation as a label (negative, neutral or positive)

• analysis-result-full.csv with full analysis results in the following columns:

  • all id columns from the input table (used as primary keys)
  • binaryData serialized data with full analysis as Base64

  This table can be used as an input for the Geneea Frida writer app.