With the development of mobile networks and intelligent terminals, the recommendation of the companion based on high-quality users has become one of the hot topics in the Internet, and the recommendation algorithm about companion is the crucial factor. In the past, the user location trajectory similarity recommendation algorithm was mainly based on geographic location or base station data and the data sparse may result in undesirable results. This paper proposes a companion recommendation model based on the cosine similarity of IP sites. More comprehensive IP sites data have been used instead of geographic data, and the date time data are calculated for cosine similarity to eliminate the data sparseness problem. Finally, the people with higher similarity and higher quality are recommended.