Data mining is increasing its popularity in the research of Technology-Enhanced Language Learning and Applied Linguistics in general. It enables a better understanding of progress, performance and possible pitfalls, which would be useful for language learners, teachers and researchers. Until recently it was an unexplored field, but it is expected to grow exponentially in the following years. This article attempts to be a relevant contribution as an instance of empirical research, showing the application of Learning Analytics to the Language MOOC (LMOOC) 'How to succeed in the English B1 Level Exam.' The focus or the research was threefold, trying to find out: (1) what types of learning objects students engage with most, (2) what aspects of online interaction relate more strongly to course completion and success, and (3) which are the most prominent student profiles in an LMOOC. Results show that short video-pills are the most powerful learning objects in this type of online courses, the regular submission of automated grading activities is a robust indicator towards course success, and the most prominent student profile in LMOOCs is 'viewers', those who access the learning materials but do not submit tasks or engage in online interaction actively, which would explain why the completion rate in LMOOCs is so low. This novel perspective into students' language learning, which big data has assisted us in, should guide course creators to re-design the LMOOC for the enhancement of the audio-visual content. LMOOC instructors and facilitators should also encourage participants to increase the submission of activities--acknowledging these small achievements through micro-credentialing and badges-, and special attention ought to be paid to the most prominent LMOOC profile, those 'viewers' who should be lured into becoming 'solvers' or, even better, 'all-rounders'.