Abstract:
This paper presents a new approach for spell-checking based on the user profile and that can be applied for any language. For this purpose and for the specific case of Arabic, spelling errors are studied and divided into 18 types. Then, a relationship model between users and their errors is obtained. The proposed architecture initially gives apposite values for a current user, then corrects misspelled words by applying the spelling rules, and the remaining words are corrected based on the probability given by an adopted model of the profile values. To show the efficiency of our profile-based approach, we conducted an experiment with a corpus of 11,908 words containing 1,888 errors. It showed that our approach suggests the correct word in 88.43% times and ranks it in the first four positions in 75.14% times. Moreover, using the same corpus we compared our implemented tool with two existing ones where ours ranked better in 69.79% times than Sahehly and 77.63% times than MS word.