Create Trending_Algorithm.MD

docs/Trending_Algorithm.MD

@@ -0,0 +1,47 @@
+# Trending Algorithm
+
+It is essential to consider specific factors in order to get the best potential result for trending detection. 
+The first thing to consider is the _increment or decrement ratio between two y-axis in a given time (x-axis)_ to detect a peak, you may use deltas _(Δf(x))_ 
+for this. 
+You would have the following mathematical model:
+
+
+_**ω** = (Δf(x)) / f(x) or **ω** = (Δy) / y_
+
+
+Where **_y_** or **_f(x)_** represents _social interaction scoring (SI)_ while **_ω_** represents the _social fluctuation ratio (SFR)_. Furthermore, SI represents 
+a _view, like, mention or comment_ in an specific aggregate. However, by using this specific ratio it is impossible to detect peaks in a _high-mass 
+environment_ such as the internet that is an Exponential-Driven and high-mass environment. SFR or ω is highly relative when comparison are being made in a high-mass environment.
+
+
+Consider the following potential scenario: aggregate "x" has 1 social interaction (SI) at the first hour (t), eventually it went from 1 to 10 SI in 
+the next hour. Furthermore, aggregate "y" has an increment of 123,980 to 125,980 SI in 1 hour (t). Henceforward, you may calculate the increment or 
+decrement ratio (social fluctuation ratio or ω) to determine if there was a peak between the given hours (t). 
+
+Using the past hypothetical scenario, 
+aggregate "x" social fluctuation ratio (ω) is 0.9 (90%) while aggregate "y" ω is 0.1587 (1.87%). Consequently, if you exclusively used this 
+mathematical model, _you won't get the result needed to calculate trends as discussed before._
+
+
+Thus, you may consider **two factors** as key to discriminate each candidate. While ω (social fluctuation ratio) works to get the SI increment and 
+decrement ratio, it lacks of a required factor to _remove the high-relativity_. Here is when the **_valuable ratio (υ)_** is mentioned. Valuable ratio 
+mathematical model is represented like this:
+
+_**υ** = f(x)/z_ or _**υ** = y/z_
+
+Where **f(x)** represents the given SI (social interaction score) and **z** represents the _maximum SI score_ in the whole data set.
+
+This **valuable ratio (υ)** gives you the _required factor to eradicate high-relativity_ in a large data set. Therefore, you may be able to calculate 
+the required score to detect a trend in a Exponential-Driven and high-mass environment.
+
+To conclude, _you must sum the given mathematical algorithms_ to get an overall score called **Trending Score** _(represented as Φ)_. It may be represented 
+in a mathematical model as this:
+
+
+_**Φ** = ω + υ_
+
+
+Where **υ** represents valuable ratio, **ω** represents _social fluctuation ratio_ and **Φ** represents _trending score_. In addition, this lends us to _store these 
+results in a specialized database_ such as a data lake to query any aggregate by it's high-cardinality field and output very specific results for each 
+scenario _(like trends for an specific user based on his consumed content and country by denormalizing aggregate fields)_.
+