Sentiment Analysis

There can be no serious sentiment analysis without deep parsing

 Like all sentiment analysis engines, the Synomia engine exploits lists of nouns, verbs, adverbs, and adjectives with positive, neutral, negative or ambiguous tonality, and also uses lists of intensifying adverbs, etc.

But all these lexical resources are of limited usefulness without the complete deep parsing of the sentences. The simple presence of positive or negative words in a sentence is not enough to evaluate if the sentence puts forth a judgment, if it is a positive or negative judgment, or to what this judgment relates. It is absolutely necessary to know the syntactic links between the words of the sentence. It is here that the Synomia engine draws all its force, and leaves standard engines lagging behind.

Two simple examples will be enough to illustrate this matter. To correctly analyze the opinion stated in the sentence “the Samsung telephone which has just come out is too expensive,” it is necessary to have located all of the following syntactic links :

  • the subject relation between the noun telephone and the verb is
  • the predicate relation between the adjective expensive and the verb is
  • the relation between the adverb too and the adjective expensive

It is in these conditions only that one can deduce not only that this sentence expresses a negative opinion (too expensive), but also that the object to which this opinion relates to is a telephone (a Samsung telephone).

The Samsung telephone which has just come out is too expensive

In the sentence “with this fiber, several screens can work simultaneously without loss of power,” there are several “inversion of tonality,” which can be understood only through syntax. Concerning fiber, power is a word with positive tone. Its syntactic dependence with the word loss reverses the tone: the nominal group loss of power has a negative tonality. This group depends syntactically on the preposition without, which depends on the verb phrase can work simultaneously, which reverses the tonality once again: without loss of power has a positive tonality.

 With this fiber several screens can work simultaneously without any loss of power

One could multiply the examples. There is a multiplicity of syntactic structures which are directly correlated with the expression of emotions. Without the syntactic analysis step, a serious analysis of emotions is difficult.

That said, even if we have, thanks to deep parsing, a sentiment analyzer more powerful than standard analyzers, we share with all serious actors of the field the belief, that a fully automated sentiment analysis is impossible, and so trying to achieve it is useless. At the macroscopic level, the results of machine analysis are positive overall, but insufficient. However the precision and wide coverage of the analyses of the Synomia engine make it possible to the company's consultants to establish diagnoses, measure strong tendencies, locate weak signals and extract relevant insights. The benefits in terms of quality and productivity are huge.




63 bis rue de Sèvres
92100 Boulogne-Billancourt

Tel : +33 (0)1 46 10 06 40

Email :

Scroll to Top