The diagram below shows the general architecture of EmoLib:

Figure 1: Architecture diagram of EmoLib.

The processing structure of EmoLib is built of several primary abstract classes that define a sequential modular framework, i.e., a pipeline, due to the dependencies in the tagging process. This pipeline is described in detail in the AffectiveTagger.

The actual (concrete) classes that perform the Natural Language Processing (NLP) tasks (identified in Figure 1 inside the Tagger block) inherit the common language-processing-wise methods and functions defined by these parent (abstract) classes. By following this specification, the system gains an additional degree of flexibility, scalability and maintainability. This responds to the need of obtaining a complete modular configurable framework for EmoLib.

The primary abstract classes that define the architecture of EmoLib are described as follows:

• Tokenizer - Splits a text string into individual units, called tokens, creating the data structure. These tokens are expressed in regular patterns as established by the grammar of the language. It is the INPUTTER of the pipeline.
• Sentence Splitter - Segments the incoming text into paragraphs and sentences.
• Part-Of-Speech Tagger - Disambiguates the function of nouns, verbs and adjectives in the sentence according to the context.
• Word Sense Disambiguator - Determines the correct sense of the words according to the context and extracts the appropriate set of synonyms.
• Stemmer - Strips the suffix of words in order to index those which share a common meaning, thus improving Information Retrieval (IR) performance.
• Emotional Keyword Spotter - Determines the emotional dimensions of the words that have an affective content.
• Statistic - Computes the averaged emotional dimensions of the words in the text.
• Classifier - Labels the text with the most appropriate affective tag according to the affective attributes.
• Affective Formatter - Presents the obtained results in a usable XML format in order to integrate it as a tool into a broader project. It is the OUTPUTTER of the pipeline.

There's also an external monitor, i.e., the Printer, to help show the information (data) that flows in the processing pipeline in order to check that the different modules that build the chain work correctly.

The configuration issues are conducted by the Configuration Manager, which receives the specific configuration from an external XML config file. This system has been taken from the Sphinx-4 speech recognition project, so please refer to the pertinent documentation on the configuration specifications at the Configuration Management for Sphinx-4.

It is believed that the availability of a simple code is the best possible documentation in order to begin using EmoLib. See EmoLib Simple Example for further details.