Hi Kohdesmond
There are a number of things that, compounded, are leading to different results.
(1) The song you uploaded looks to be only 30.079 seconds long (as reported in the audio summary of the track profile call). So it looks like you've only uploaded a 30 second sample, whereas the full track is three minutes long. The audio summary of a 30 sec. snippet of a song will always be different than the summary for the full song. The energy and valence could be different depending on which 30 seconds was analyzed. In this song, for instance, the break at 2 minutes has a very different acoustic profile than the beginning or end. Also our analysis will naturally give more accurate data when you give it more input data, so the results from the full song will always be better than results generated by analyzing a 30 second snippet.
(2) The two songs might actually be different versions of the same song. There are many remixes, acoustic versions and such for the Jackson 5 classic. It is quite possible that you are seeing results from different versions of the song and thus will get different results.
(3) The songs have been analyzed with a different version of the analyzer. Over time, we make updates and improvements to the analyzer, so you may see values for parameters like the energy and valence change when we make an update. The detailed analyzer data shows the version used, if they are different, that would explain why some results may be different.
If you want the most accurate results for an MP3 and you have the audio, then the best approach is to analyze the whole track, not just a 30 second sample. Of course, that takes a lot of time and bandwidth, especially if you are working with lots of songs. For many apps, getting the summary based on a metadata match is quite adequate.
Hope this helps
Paul