In what is probably his most famous poem—Poema 20—the Nobel Prize-winning poet Pablo Neruda wrote: “Es tan corto el amor, y tan largo el olvido." (Love is too short, and forgetting too long.) Most people relate to Neruda's words. When in love, people cannot help being reminded of their love interests. But, when love fades, these memories become more distant, and less recurrent, as they fade into oblivion.
Borrowing inspiration from Neruda's words we can say that societies also fall in “love,” when they experience periods of intense media attention targeted at a new celebrity, recently elected leader, or a new piece of cultural content, such as a new movie, book, or song. But does society also suffer from Neruda's predicament? Is society also intense in “love” and slow in forgetting? Indeed, recent work suggests that the decay of a cultural piece should be characterized by an initial period of intense attention, followed by a slow forgetting. However, the decay or forgetting of cultural information has been little studied.
In this work, we use data from Spotify, Last.fm, YouTube, Wikipedia, IMDB, and Billboard Magazine, to show that the online attention received by each piece of cultural content decays as a function of its age following a two-step exponential function. We also show that a simple model, motivated by the vast theoretical literature on memory studies correctly reproduces the observed behavior; with an initial fast decay due to the fast rate at which cultural pieces decay from our communicative memory, followed by a relatively slower decay due to the slower rate at which cultural pieces decay from our cultural memory.
We model the attention received by a cultural piece by assuming that when individuals decide to consume a piece of content, they will sample it from a pool of collective memory proportional to the age of each piece of content. As a measure of collective memory, we consider the relative size (S) that each cultural piece of age t takes in our collective memory, it can be separated in two, one part from communicative memory (v) and another from cultural memory (u). And both decay, with different rates (p and q respectively), on age (Fig. 1).
Our model allows us to estimate the moment in which the communicative memory does not matter anymore, and all the forgetting is due to the information decaying from our cultural memory. For example, we find that songs remain for 5 years in average in our communicative memory, movies remain 11 years in average and cultural icons like athletes remain between 20 and 30 years depending on the discipline (Table 1).
Besides, we find initial performance matters. For example, for songs, we find that the highest peak and the number of weeks into the billboard ranking are good predictors of current popularity. In particular, we find that the current popularity of a song can be explained by its age and its initial performance with 60% of accuracy.
Our findings show that the dynamics of human forgetting is quite universal (Fig. 2) and it can be characterized by a narrow set of mathematical functions and that the theories about collective memory advanced in previous literature can be formalized in a mathematical model that can capture the empirically observed forgetting dynamics.