Search other topycs

Categories

The ideal influencer for my brand

Artificial intelligence for social networks

In the current state of the influencer market, the gold standard is to choose an influencer based on their volume of followers. In this work we show that this criterion, as the sole decision factor, is insufficient to adequately evaluate the influencer. Influencer marketing is the strategy that companies use to advertise in collaboration with well-known people on social networks. Through publications, recommendations and testimonials, influencers suggest buying behavior on a specific market, highlighting a brand and promoting sales.

Transmit the brand

However, it is not enough to find a person who publishes about the product, it must inspire confidence in the audience and be a reference in the subject they are talking about. For example, companies like Adidas or Yanbal, hire figures with a significant fan base, such as Paulina Vega (@paulinavegadiep) or Ita María (@ itamaria83). These influencers are not chosen at random, Paulina Vega is recognized for being Miss Universe 2014, an achievement that could hardly be achieved without maintaining a healthy lifestyle, which makes her a perfect candidate to support the sports brand.

Paulina Vega recomienda adidas
Paulina Vega’s publication where she recommends an Adidas “look” to exercise.

On the other hand, as fashion and makeup go hand in hand, Yanbal chose Ita María as one of her brand’s ambassadors, who calls herself a “Fashion Insider”.

Ita Maria promotes Yanbal
Publication of Ita María where he promotes a Yanbal contest.

Is it the ideal influencer for the brand?

Measuring the effectiveness of an influencer is not easy. In different studies such as the one elaborated by Cha et al. (2010), influential people can be characterized according to metrics such as number of followers, mentions and retweets. The influence is given by the value of the user (mentions) and the value of their messages (retweets). They also characterize the most influential celebrities for being more mentioned, the media for having more followers and content aggregators for being more retweeted.

Likewise, there is a line of studies such as that of Segev (2018), which tries to validate the veracity of social media metrics using regression models to predict the scope or number of views of a publication given the number of comments, followers and likes. The authors conclude that follower volume alone is not a good indication of reach, compared to other interaction metrics such as likes and comments.

The best way to study an influencer is to consider as many metrics as possible. However, the conventional dimensions of social networks can give biased results, since they do not take into account the sentiment of the comments and there is no way to identify the veracity of them.

For example, a user who has a lot of followers, likes and comments would probably be identified as a good influencer; But if the comments reflect a negative feeling towards the character, it can go against the business objectives.

Natural Language Processing

Due to the growth of influencer marketing, a huge market has emerged around social media. It is now possible to buy likes, followers and comments, triggering a big problem, where the user seems to be a better influencer than he really is.

Considering this situation, at Grupodot we created a product that directly attacks the first problem, allowing influencers to be characterized, not only using traditional metrics, but incorporating sentiment analysis through Natural Language Processing (NLP).

Natural Language Processing

With this tool it is possible to analyze the performance of different influencers on Instagram, so that it is easy to identify which ones are most appropriate for the identity of the brand or the intention of the campaign.

By integrating traditional indicators such as the number of followers, along with sentiment analysis, on a volume of texts related to an influencer, it is possible to quantify the reputation of the influencer, in terms of feelings, associated with the comments he receives from his followers.

Influencer analysis

For the analysis of influencers, we built a scraping tool that extracts data from social networks such as Instagram and uses the Natural Language Processing (NLP) Google API, to perform sentiment analysis on the comments associated with important publications of the selected influencers, according to the client’s need.

From the results of the sentiment analysis we can obtain:

  1. Performance metrics for each publication.
  2. A ranking of influencers, based on the metrics for each one.
  3. Visualizations that allow a clear and concrete understanding of the global performance of each one.

Analysis of an Instagram post

Using the scraping technique, we automatically extract all the comments related to a certain number of publications from each influencer. Thus, for each publication we obtain the following attributes (Figure 1):

  • Total number of likes
  • Total number of comments
  • Full text of each comment
Data extracted from publication
Figure 1. Example of data that is extracted from a post on Instagram. (The number of comments is not visible directly on the publication’s web page, but in the html code of the publication).

The scraping process is structured in such a way that two types of tables are obtained:

  • A table with the total likes and comments for each publication and influencer (Table 1).
  • A set of tables for each influencer, where each table stores the detail of all the comments of a publication (Table 2).
Total likes and comments from influencers
Table 1. Example of a table with total likes and comments obtained by publication and influencer.
Comments extracted from publication
Table 2. Example of text detail of the comments extracted from a publication.

At this point, it is important to mention that each time the API is used to analyze a text block, we obtain as a result two numbers called score and magnitude, which are interpreted as a measure of positivity (score> 0), negativity (score <0) and text size (magnitude> 0), respectively.

In our case, these numbers are calculated for each comment and based on the results, we define unique sentiment indicators per publication. So, to measure the performance of a particular publication, we define the PPS or Post Sentiment Score (m0). This metric is a weighted average of the sentiments of all the comments in that publication, where the scores are averaged using the magnitudes as weights:

Post sentiment score analysis

To better understand these results, a scatterplot can be made of all the comments on the publication, where the axis of the publication are precisely the score and magnitude. The result is a map that allows us to see the sentiment distribution of all the comments of a publication (figure 2).

Analysis sentiment map

Then, counts are made to obtain the total of positive, negative and neutral comments per publication, building a table with the metrics of each one. (table 3):

Table of misgivings and results in a publication
Table 3. Example of a table with sentiment results per publication: The red box shows the magnitudes contained in the definition of the PSS and the blue box shows the sentiment centroids for each publication.

Analysis of influencers on Instagram

The results in Table 3 are grouped by influencer, to calculate a metric that allows each to be characterized with a unique score. For this, a weighted average of the PSS of the recent publications of the influencer is considered, this metric considers that the weights are a sum of terms proportional to the number of likes, comments and views:

Weighting metric likes comments and views

Here we consider some relevant percentages according to the objective of the campaign (reach or engagement). We call the score calculated by influencer IGM (Influencer Generalized Metric):

Influencer Generalized Metric engagement

On the other hand, it is interesting to consider a metric that only involves the sentiment generated by the publications of an influencer, regardless of the volume of followers that he has, (to evaluate micro-influencers). Therefore, it is useful to consider the average PSS of all the publications of the influencer as another additional metric, which we call MSS (Mean Sentiment Score).

Using these formulas, we build ranking tables considering several renowned influencers in the country:

Tablas ranking influencer

In both cases, the influencer number 1 is @jbalvin, thanks to its large number of followers, the IGM weights were chosen to give greater importance to the volume of followers. Thus, if one wants to have scores that reflect the effect of the other indicators, it would suffice to modify the percentages within the IGM.

We also see that to have a more accurate ranking, it is necessary to create categories of influencers separated by quartiles, after this, carry out another ranking calculation for each category: The IGM normalizes likes, comments and views with the medians over the entire set of influencers . This means that these metrics for the same influencer, vary according to the other influencers with whom it is being compared. In this way, comparing @greeicy, @la_segura, @el_mindo and @paulinavegadiep without considering @jbalvin can be much more accurate and precise.

Another way to visualize comparisons between influencers is to use spider diagrams to show what are the strengths of each: If we now compare the sentiment histograms of the top 5 of our ranking:

Which is the most indicated?

Vemos que @jbalvin no tiene las publicaciones con el mejor sentimiento, en su lugar, resulta que @greeicy1 y @paulinavegadiep son quienes muestran una conexión más positiva con su influenciadores. Esto puede sugerir que las publicaciones contenidas en ese grupo de puntajes super positivos, poseen un tipo de mensaje muy diferente al de las demás publicaciones de ese influenciador.

Another way to visualize comparisons between influencers is to use spider diagrams to show what are the strengths of each: If we now compare the sentiment histograms of the top 5 of our ranking:

Influencers web diagram

Which is the most indicated?

In this diagram we consider the variables variables diagrama  which represent the averages of comments, likes and the total of negative, neutral and positive comments, respectively. With this example we observe that:

  • The influencers who receive the most likes do not necessarily receive the most comments. Within the top 5, everyone has a tendency for comments to be positive.
  • Within the top 5, everyone has a tendency for comments to be positive.

Here we can analyze that if you want to achieve enough likes with the campaigns, @ greeicy1 is the most suitable influencer, on the other hand, if the objective is to get people to participate by leaving their comments, then @la_segura is more appropriate.

It is curious to note that @jbalvin is not the leader in either of those two metrics. variables diagrama which shows that an influencer with a large number of followers is not necessarily going to achieve the reach or engagement that a certain advertising campaign requires.

This is a deeper exploration of interactions with followers, in terms of various metrics such as the number of likes and comments, and text analysis using natural language processing algorithms to determine sentiment, offering more accurate indicators of reputation. of the influencer within its audience volume.

Below we highlight the following advances:

  • A standard index was created that incorporates traditional metrics (likes and comments) and the sentiment generated by an influencer.
  • Through visualizations such as sentiment histograms and spider diagrams, it can be seen that there are influencers with better sentiment or higher engagement, even if they have more little audiences.
influencers engagement

To highlight the strengths of each, influencers can be classified into two groups: those with the most interaction (with the most comments) and those with the greatest reach (most likes).

However, in order to more accurately determine the efficiency and impact of an influencer, it is pertinent to develop a false metric detection system.

In this way, the influencer’s score will be much more reliable, allowing companies to be informed to guarantee a greater return on their investment.

In the world of advertising, Instagram is a social network that is considered relevant among those users who are famous personalities or who have a very large fan base. On the other hand, Twitter is a social network that, while popular on political issues, is preferred by many micro-influencers on advertising issues. In a next installment, we will show how to do advanced analytics to characterize micro-influencers on Twitter.