Abstract: Recommendation system algorithm is one of the important technologies of big data mining. In this study, we applied the content-based recommendation system algorithm to construct an utility matrix of the active and factor items, based on data obtained from the 1∶50000 geological and mineral resource survey at the Wendi district of southern Qinzhou-Hangzhou metallogenic belt, South China. The predicted active item included the middle- and small-sized (ore spot) electrum and verified non-electrum deposits; it also included the unevaluated electrum deposits, medium- and small-sized (ore spot) lead-zinc and verified non-lead-zinc (ore spot) deposits, and the unevaluated lead-zinc deposits. Caledonian migmatite, early Yanshanian and late Yanshanian intrusions, NE- and NW-trending faults, and Au, Ag, Pb and Zn elements were considered the factor item. The Euclidean distance similarity between known deposit (or ore spot) and other unevaluated areas was calculated and then used to predict the prospecting area of silver-gold and lead-zinc deposit. The results show that the recommendation system algorithm can effectively mine mineralization related information, and quickly extract the potential deposit areas based on its similarity to certain types of deposits (ore spot). For electrum deposit within the Wendi district, high similarity areas were mainly distribute around known ore deposits and on both sides of NE-trending faults, with a small portion distributed near the overlapped fracture. In comparison, for lead-zinc deposits, medium-sized deposit showed a high degree of discrimination. The high-value areas covered almost all known lead-zinc deposits while more concentrated distribution were found for small-sized deposits. In addition to known deposits, several high value areas can also be used as key prospecting targets.

Key words: big data mining, recommendation system algorithm, utility matrix, ore deposit forecast, Qinzhou-Hangzhou metallogenic belt, Guangxi Wendi district