The 7 Unique Insights

  • Association relationships are bi-directional: if two features are associated, it means they are interlinked and you can replace one with the other.

    Having a strong association means that any time an event occurs with one feature it will most likely occur with the other.

    It also means that associated features happen together, irrespective of whether or not they occur quite often or rarely. They therefore tend to always happen together.

  • Influencer relationships are one-directional: if one feature is an influencer on another, it means they are linked in a simple direction, going from the influencer to the target.

    Having a strong influencer relationship means that any time an event occurs with an influencer, it will most likely occur within the target (which is being influenced as a result).

    It also means that influence is always happening, irrespective of whether an event happens often or rarely.

  • If one feature is a cause to another, it means that whenever the cause occurs, the effect will occur with high certainty.

    If one feature is not a cause to another, it means that whenever the cause occurs, the effect will not occur with high certainty.

    Having a strong causal relationship means that any time a cause event occurs, it will most likely lead the effect to occur.

    It also means that the cause is consistently active and capable of causing (or not causing) its effect across all situations.

GoFlek brings its own IP, with proprietary algorithms and formulas to uncover essential relationships among variables in datasets.

  • ANOMALY :

    If one feature is an anomaly to another, it means that whenever the anomaly feature occurs (which should be rare), the target must occur.

    Having a strong anomaly relation means that any time a rare event occurs, it will most likely affect the target to occur.

    POLYMALY :

    If one features is a polymaly to another, it means that whenever the polymaly feature occurs (which should be very common), the target must occur.

    Having a strong polymaly relation means that any time a very common event occurs, it will most likely affect the target to occur.

  • To make a prediction, you need to run a series of algorithms to ensure a correct forecast is made.

    With GoFlek, all predictions are traceable and you can peek into how the prediction score was calculated.

    Having a high prediction score means that any time the input variables occur, they will most likely lead the target to occur.

    Unique algorithms, based on a series of multiple probabilistic measures like: Joint, Conditional, Naive Bayes, Look-A-Like, Most Frequent, Latent Cosine, Jaccard, Lift.

  • Allows running both item-based and profile-based recommendation using probabilistic formulas.

    With GoFlek, all recommendations are traceable and you can peek into how the recommendation score was calculated.

    Having a high recommendation score means that any time the input variables occur (which could be items or profile, or both) they will most likely lead the target to occur (be recommended).

    Unique Algorithms based on a series of multiple probabilistic measures like: Joint, Conditional, Naive Bayes, Look-A-Like, Most Frequent, Latent Cosine, Jaccard, Lift.

What makes GoFlek’s insights different from other methods?

GoFlek’s insights go beyond the usual probabilistic measures and data science techniques.

  • Today there are two methods to measure association:

    • Correlation, which is based on associating two numerical variables based on their statistical similarity — they either increase or decrease together, or one increases while the other decreases.

    • Probabilistic, which is based on using joint probability as an association measure.

    Why is GoFlek's measure a good measure of association?

    • We know that correlation only works with numerical variables — what if one variable is numeric and the other is categorical? What if both are categorical?

    • We also know that joint probability is a weak measure of association, which becomes clear when, for example, two variables have a low joint probability yet occur together all the time (strongly associated).

  • Today there are two methods to measure influence:

    • Regression, which is based on finding a formula that fits two or more numerical variables — where a high coefficient indicates a variable has strong influence.

    • Probabilistic, which is based on using conditional probability as an influence measure.

    Why is GoFlek's measure a good measure of influence?

    • We know that regression only works with numerical variables — what if one variable is numerical and the other is categorical? What if both are categorical?

    • We know that conditional probability is a weak measure of propensity to influence, which becomes clear when, for example, the target has a high probability and the influence (condition) has a low probability, yet Pr(target/influence) ≈ 1.0.

  • Today there is a common method to measure causality:

    • Causal Bayesian Network, which represents a probabilistic graph (DAG) with conditional probabilities, with specific assumptions on the independence of variables and overall probability distribution.

    Why is GoFlek's measure a good measure of causality?

    Finding data that can be represented as a DAG and then measuring causality is prone to many unrealistic assumptions and human interpretation. Additionally, it is based on the independence of variables, which places a heavy and unnecessary condition on the relations in the DAG.

    In view of the above limitations, GoFlek's measure does not attempt to fit the data into a DAG and does not make any assumptions on the independence between variables.

    Furthermore, it measures causality in both directions (0 to 1 and 0 to -1), since causality can have both positive and negative effects.

    Moreover, GoFlek's causality measure is context-sensitive — meaning it can be measured in different contexts (whether or not other variables are taken into consideration).

  • ANOMALY

    Today there is a common method to find anomalies: Search for rare combinations of events that seldom occur (probabilistically or statistically rare).

    With GoFlek, anomaly takes a more realistic meaning with an added condition. Having a rare event is not the only condition for an anomaly; the additional condition is that this rare event must lead to the occurrence of the target event being measured.

    For example, if a person entering a shop has blue eyes, green hair, and dark skin, it is a rare event — this event is insignificant unless it is related to the fact that this person made a purchase.

    Furthermore, this rare event must be recurring, in the sense that a person with this profile, whenever they enter the store, makes a purchase.

    POLYMALY

    Today, there is no concept of polymaly in literature; this is an innovation of GoFlek. Measuring polymaly is quite important.

    With GoFlek, polymaly allows the detection of a very important pattern of common events that lead to the occurrence of the target event being measured.

    For example, if a person visiting a website reads a specific article in the blog section — this event is insignificant unless it is related to the fact that the same person always fills out the sales request form.

    For this event to become a polymaly, it must be recurring quite often (meaning that many people read this specific blog post), and whenever they do, they convert.

  • Today there are common methods to make predictions:

    • Classification: Using decision trees to categorize data into predefined classes.

    • Regression: Estimating continuous numerical values, such as future stock prices or house values.

    • Time Series: Analyzing data sequences over time to identify trends, such as monthly revenue or seasonal patient admission rates.

    Making a prediction with GoFlek is different in the sense that it runs a series of probabilistic algorithms to ensure a correct prediction is made — the search is exhaustive and recursive, with all combinations explored until a match is found. Moreover, the prediction is traceable and users can peek into how the score was obtained.

  • Today there is a common method to make recommendation:

    • Collaborative Filtering which recommends items based on the behavior of similar users (e.g., "users who bought this also bought that")

    • Content-Based Filtering: Recommends items by comparing the item's profiles (e.g., genre, director, description) to the user's past consumption history and preferences.

    • Hybrid: Takes both techniques and meshes them together.

    To make a recommendation with GoFlek users are able to do both Item-based recommendation, Profile-based or both using ONE single algorithm and using Probability as a base.

    Our Recommendation is traceable and users can peek into how the score was obtained. Also, GoFlek Recommendation is fast because it is based on simple probabilistic algorithms compared to Collaborative filtering which is computationally expesive (because it is based on complex Matrix operations).