In this paper, we applied a peer assessment scenario at the Technical University of Manabí (Ecuador). Students and professors evaluated some works through rubrics, assigned a numerical score, and provided textual feedback grounding why such a numerical score was determined, to detect inaccuracy between both assessments. The proposed model uses soft computing techniques to reduce the professor's workload in the correction process. Experiments were carried out with a data set in the Spanish language. We applied a supervised machine learning approach to obtain a sentiment score corresponding to specific textual feedback, and the fuzzy logic approach to detect inaccuracy between numerical and sentiment scores and obtain the assessment score. The results showed that the support vector machine model had a better performance with low computational costs when the feedback was represented as a 1-g and 2-g vector, whose relevance was weighted with term frequency-inverse document frequency; moreover, the grader's critical judgment validity was inferred from the similarities between numerical and sentiment scores. At the end, the outcomes assert the model is reliable and guarantees a fair peer assessment procedure.