M. Buelow 01.09.2023, translated from German by ChatGPT 3.5
Words are associated, in a purely formal context, to a certain extent, with positive or negative emotions. This is the positive or negative connotation of a word. If it has been determined for a text how often certain words function as subjects and how often they function as objects in sentence structure, this comes close to the word's connotation. Subjects are more positively connotated, while objects are more negatively connotated. Connotation can thus be calculated through such a tally: Number of occurrences as subject minus number of occurrences as object. The larger this value is in the positive range, the more positive the emotional tone for the word; a negative value indicates a negative tone, with the magnitude representing the intensity. Generally, a linguistic opposition always consists of a positive and a negative term. However, not every opposition is considered here; only the conceptual difference between subject and object is taken into account. In a speech, for example, the subject embodies positive ethos (Who is speaking?) and pathos (How is it delivered?), influencing pragma (What is being presented?). The object, on the other hand, merely serves as the contrary reference point that is typically not active. The determined emotion for a word is strictly speaking a specific one, based on the difference between subject and object – the so to speak subject-object-emotion. It yields decent results for the general connotation of words. In this process, the focus is on sentence structure, i.e., the actual content of subject and object.
Sentiment analysis from unlabeled data can recognize the emotions associated with words for any language without prior preparation. In the above approach, there is no need to create a training dataset with solutions, such as training a neural network. The sentiment is learned unsupervisedly and automatically from an original text. Each text serves as the basis for learning the sentiment associated with its own words. This sentiment is the one generated within the text. Furthermore, if one stores the positive or negative connotation as an average across multiple texts for individual words in a database, the difference from the current text indicates the sentiment trend. It represents the overall sentiment shift conveyed by the current text in relation to the norm of individual words.
The open-source application mbFXWords utilizes OpenNLP for text parsing into sentences and the determination of subject, predicate, and object in each main sentence or subordinate clause. Essentially, the application takes advantage of the fact that subject, predicate, and object (S-P-O) appear in this sequence within a sentence: In the class OpenNLP.java: The distinction between subject, predicate, and object is made towards the very end of the listing. The freeware application ShelfClauses additionally tallies the occurrences of each word as a subject minus those as an object, thereby determining sentiment. This sentiment is indicated by the red (negative sentiment) or yellow (positive sentiment) highlighting of words in the ShelfClauses word table. The intensity of sentiment is represented by the darkness of the red color or the brightness of the yellow color.
to main page