While online advertising usually incorporates eye-catching images or pop-ups to attract user attention, it is less likely to mine data generated by user interaction with the system when browsing. This study attempts to utilize natural human reading behavior and determine which contents user attention is focused on to investigate users’ underlying preferences and recommend targeted advertising for said users. The experiment utilizes eye tracking technology combined with Venn diagram concepts to propose a data mining method. Using Venn diagram concepts, words in different sets are assigned weighted values and are transformed into units that can be used for calculation. This study first verified the effectiveness of the method of categorization, and then analyzed its predictive ability via experimentation. Results show that, in terms of advertising recommendation predictivity, the system exceeds the value of the threshold for non-random probability.