This paper is a corpus-based method for extracting users’ explicit preferences from free-text part of the registered user profile in Indian matchmaking system using Second order Co-occurrence PMI (SOC-PMI).In online Indian matrimonial system, while registering, users are asked to provide information about themselves, family and the qualities he/she is looking for in a desired partner but there is no means to automatically prioritize the desired partner features (which features are more important to him/her and it varies from person to person) except for explicitly asking the user. Extraction of users’ preferences from unconstrained attributes has not been explored yet. This motivates us to extract users’ preferences from an unexplored area. The contribution of this paper is focused on the above gap found in the research. The methodology explained in this paper automatically prioritizes these features which can be used to design a Weighted Reciprocal Recommendation model to generate more efficient recommendations. Experimental results show the efficiency of the applied methodology.