Exploring the use of gender-fair language by influencers

1. Introduction

Gender Equality (GE) is a significant problem worldwide, and the current pandemic jeopardizes many achievements in health, economy, safety and welfare (UNFPA, 2020). Previous research about Gender-Fair Language (GFL) has established that language is a variable that influences the attitudes, behavior and perceptions of society (De Lemus and Estevan-Reina, 2021; Patev et al., 2019; Heilman and Caleo, 2018; Sczesny et al., 2016; Koeser et al., 2015). Other studies show that GFL aims to reduce gender stereotypes and discrimination (Sczesny et al., 2016; Koeser et al., 2015), which is now well established. The measurement of the use of inclusive language (IL) by both Private and Governmental Influencers will assist in improving organizational policies.

The motivation of this study is to contribute to the achievement of Sustainable Development Goal Number 5, Gender Equality (GE), from Social Marketing (SM). In this context, defining DMs for the use of IL becomes relevant to the new consumption of marketing technology (MarTech), i.e. more awareness, inclusion, and a social sense of welcoming and belonging. Defining an inclusive marketing (IM) with DMs implies working with the conviction that the DM communication strategies must promote a social value that requires a change in attitude and behavior throughout society (Lin et al., 2021).

Research on the use of DM and MarTech tools for Decision Support Systems are topics that have been growing in the literature (Humphrey et al., 2021). Studies on the topics now reveal the benefits of using MarTech for performance management related to Customer Experience analysis. Artificial Intelligence (AI), predictive and automated learning analyses incorporate into DM (Kotler, 2021).

The questions research hypothesis as this study answers are: (1) What is the level of use of IL by the Top Instagram Influencers in Chile in 2021? (2) What words allow for determining the minimum degree of use of GFL? (3) What is the relationship between the Top Instagram Influencers in Chile in 2021 and the use of IL in DM?

The study aims to explore the use of GFL by Influencers on Instagram by proposing a Bag-of-Words (BoW) Method called GFL Clustering BoW Methodology to identify whether an IM strategy can be used. Thus, this research has a methodological and practical contribution to increasing the number of MarTech tools.

In practice, this study will help define communication strategies for DMs in public and private organizations to improve the identification of IC in social networks (SN) and can contribute to a more Inclusive Society. In addition, the study provides a methodology that supports teams creating DM by indicating which aspects of Instagram posts may generate stronger or weaker reactions from users, what being an Influencer means in terms of interactions, likes or comments per post, and what this implies in terms of engagement.

The rest of this paper is organized as follows: (2) Theoretical background, (3) Methodology, (4) Implementation of the methodology based on technological architecture, (5) Conclusions, (6) Managerial implications and (7) Limitations and potential future research directions.

2. Theoretical background

This study aims to explore the ways and degree of use of IL by Influencers in their SN (i.e. Instagram) to propose a methodology for digital content (DC) analysis as a tool that, in turn, supports SM analysis. Previous research shows how language is a mediator for the achievement of social inclusion (SI) and social diversity (Tankosić et al., 2021; Piller and Takahashi, 2011). In addition, other authors have established that language is a variable that influences the attitudes, behavior and perceptions of society (De Lemus and Estevan-Reina, 2021; Patev et al., 2019; Heilman and Caleo, 2018; Sczesny et al., 2016; Koeser et al., 2015; Koeser and Sczesny, 2014). The literature has identified a gap in the research of more complex methodologies based on SN information (Nicolas et al., 2018); therefore, this study employs Text Analysis by Data Mining Algorithms (Urrutia et al., 2021) to develop the GFL Clustering BoW Methodology.

3. Methodology

The GFL Clustering Methodology proposed uses a BoW of IL and non-sexist words (NSW) that we call “ILandNSW language”. It may apply to Influencers of any country, not only on Instagram but any other SN like Facebook, YouTube, Twitter, TikTok or others. It has the following stages: (1) Influencer Identification, (2) Data Extraction through a Web Scraping (Landers et al., 2016) code, (3) Transforming and Loading with the Pentaho Data Integration (Li et al., 2021) tool, (4) Language Analysis of a Target Group using IL and NSW written by Influencers on their posts and (5) Data Visualization using the Power BI tool.

This Instagram Clustering Methodology can help

1. To generate a Data Dictionary with a BoW of the Target Group using ILandNSW.

2. To identify Conversation Nodes of Influencers and categorize them by the Types of Influencers.

3. To identify whether a DC by an Influencer is complying with the ILandNSW of the Target Group.

4. To obtain a Data Extraction process.

5. To generate a Platform to visualize the Data Extraction process.

The following Set of Stages analyzes Instagram Influencers of any country:

• Stage 1: Identifying and Selecting the Influencer Target Group.

• Stage 2: Generating a GFL BoW for IL and NSW used by Influencers on Instagram.

• Stage 3: Identifying and Extracting Profile and Instagram Data with Web Scraping.

• Stage 4: Applying an ETL process with Web Scraping.

• Stage 5: Creating, writing and saving the Profile and Instagram Data.

• Stage 6: Analyzing the GFL by the Influencer Target Group.

• Stage 7: Visualizing the Dashboard generated on Power BI.

Scopes: Concerning the BoW, this was created before the text analysis stage and included words belonging to the Influencer Target Group ILandNSW language. This BoW may grow if new words used in the ILandNSW language studied are discovered.

Concerning the BoW, this was created before the text analysis stage and included a small group of Spanish words belonging to the Influencer Target Group ILandNSW language used in Chile. This BoW may grow if new words used in the ILandNSW language studied are discovered. Adding new lines to the Excel file that stores the BoW and also adding other languages (English, French and so on), creating a new Excel file that stores other languages. Any difficulties you may encounter when doing something similar in another language are resolved by creating a new Excel file that stores the required words in other languages.

4. Implementation of the clustering methodology based on technology architecture

In this section, the results for the application of each proposed stage are shown in detail and the implemented Web Scraping code and dashboard are presented. The results of the same allow for answering the research questions of the study.

• Stage 1: Identifying and Selecting the Influencer Target Group

In this stage, www.starngage.com shows the Top Instagram Influencers by country and year. In this study, the Country is Chile and the Year is 2020.

The Target Group is the first ten Chilean Influencers. Table 1 shows the Instagram Data. The parameters considered are Instagram Influencer Name, Followers, Following, Avg. Likes, Avg. Comments, Participation Rate and Total Publication (Posts).

Among these Influencers of the Target Group are Writers, Athletes and people from Entertainment and TV, all classified as Influencers based on their Number of Followers.

As shown in Figure 2, Arturo Vidal (@kingarturo23oficial) is the Influencer with the most significant Number of Followers, with a total of approximately 14 million, corresponding to 34.8% of Total Followers. In contrast, the Influencer with the least Followers is Raquel Calderón, with around two million, corresponding to 4.14% of Total Followers.

As shown in Figure 3, Accounting has 30% of the Total Followers. Three Influencers have this occupation. The occupations with the least Influencers are Lawyers, Singers and Professional Soccer Players, with one User each corresponding to 10% for each Influencer.

The Number of Followers by the Influencer and The Number of Influencers by Occupation is a Pie Chart Power BI.

• Stage 2: Generating a GFL BoW for IL and NSW used by Influencers on Instagram

In this stage, an Excel file stores the BoW that contains the IL and NSW. The column “Palabras” (Figure 4) is called “ILandNSW_words” in Stage 6.

The Identification of ILandNSW is conducted manually from Literature and comments from

Instagram, Facebook and Twitter. Other sources are websites, TV and Newspapers, among others.

• Stage 3: Identifying and Extracting Profile and Instagram Data with Web Scraping [3][4].

In this stage, Web Scraping will get the Instagram profile, such as the Instagram Influencer name, publication ID, publication description (what the Instagram Influencer wrote on the post), the number of comments and the number of likes per post (publications) for each Influencer of the Target Group.

To get a public Instagram profile, a specific JSON (JavaScript Object Notation) URL [5] is implemented, see Figure 4.

The format is

profile = https://www.instagram.com/<instagram_influencer_name>/?__a=1.

In other words, to access a public influencer feed, the URL has a unique < Instagram_influencer_name > without the parameter @ that characterizes usernames on Instagram, and the parameter?_a = 1 to read all the Instagram Influencer posts. Table 2 shows the Instagram Influencer Name and the unique JSON URL for each Influencer of the Target Group.

• Stage 4: Applying an ETL process with Web Scrapping

In this stage, the JSON library, REQUESTS library and CSV (Código Seguro de Verificación) library extract, transform and load the profile and Instagram data for each Influencer of the Target Group using Web Scrapping. The JSON format structure to browse the “Instagram_data_not_clean” file is shown in Figure 5.

The Python Code (Figure 6) starts with import json, import requests and import csv. The array tags contain the Instagram Influencer names (see Table 2), and the array keys contain the Instagram profile.

The Instagram profile extraction from the JSON file considers the following parameters: id (Publication ID), owner (Instagram Influencer name), edge_media_to_caption [6] (text), edge_media_to_comment [7] (count) and edge_like_by [8] (count).

The Instagram data extraction from the JSON file considers the following parameters: “graphql”, “user”, “edge_owner_to_timeline_media”, “edges” and “node”.

The WRITER function csv.writer(Instagram_data_not_clean) [9][10] converts the Instagram data into a delimited strings chain with a write( ) method that writes on the CSV[11] called “Instagram_data_not_clean.csv”. Therefore, once the file is created or opened, the function csv.writer( ) is employed to deliver a writer object that transforms user data into a delimited chain of characters.

The function FOR for tag in tags fetches the Instagram profile and the Instagram data from the JSON file writer = csv.writer(“Instagram_data_not_clean”) for each Influencer of the Target Group through the array tags that sweep the tag and extract each url of Table 2, one by one through the function get_ig_page(url) to then use them in the function “fetches” through the variable ig_data_dict.

The function def get_ig_page(url, session = None) contains two parameters, “url” that returns the content of the page, and “session” [12] that calls the requests.session( ) method and calls the requests GET method session.get(url) that extracts the value of the variable “url”, i.e. the content of the Influencer page, and saves it in the variable “response”.

The “response.status_code” [13] method saves the value of the status in the variable “response_status_code”.

In the function IF if response_status_code =  = requests.codes.ok [14]. If the value of the requests status code is equal to the value of the status code, the value of the variable “url” is assigned to the variable “response”; otherwise, the value of “response” remains as None.

The function def fetches(ig_data_dict, writer) contains two parameters, “ig_data_dict” the content of the page and “writer” the JSON file [15]. This function has two IF and one FOR.

In the function IF if ig_data_dict is not None, it calls ig_data_dict.json( ) method that completes the data crawled with graphql keyword.

The requests GET method ig_data_dict.get(“graphql”, None), data.get(“user”) and user.get(“edge_owner_to_timeline_media”, None) extract the values of the attributes “graphql”, “user” and “edge_owner_to_timeline_media”, which contains an Instagram data fetch parsing, an Instagram Influencer name, and a photo or video. Those values are stored in graphql, user and posts, respectively.

In the function IF if posts is not None, it calls the requests GET method posts.get(“edges”,None). To browse “edges”, each publication for each Influencer of the Target Group (see Table 2) is contained by a different “node”. The Instagram profile extraction starts once inside the “node”.

To navigate the posts (publications for each Influencer of the Target Group), the function FOR for post in posts accesses the node for each post iteration.

The variables to grab are id, owner, edge_media_to_caption, edge_media_to_comment and edge_liked_by. In other words, the Publication ID, the Instagram Influencer name, the description of the photo or video posted by the Influencer on Instagram, the number of User comments per post, and the number of User likes per post for each Influencer of the Target Group (see Table 2), respectively.

Below show the Instagram profile extraction process scraping = post.get(“node”, None).

Instagram Scraping for Id: The function IF if keys =  = “id”, a variable id stores the Publication ID for each publication. The requests GET method scraping.get(keys) extracts the value of the array keys. The dumps JSON method json.dumps(keys) decodes the value and saves it in the variable id, as shown in Figure 7.

Instagram Scraping for Owner: The function IF if keys =  = “owner”, a variable owner stores the Instagram Influencer name (see Table 2). The requests GET methods scraping.get(keys) and keys.get(“instagraminfluencername”) extract the values of the array keys and the Instagram Influencer name “instagraminfluencername” in the study. The dumps JSON method json.dumps(instagraminfluencername) decodes the value and saves it in the variable owner, as shown in Figure 8.

Instagram Scraping for Edge_media_to_caption: The function IF if keys =  = “edge_media_to_caption”, a variable edge_media_to_caption stores the description per post of photos or videos posted by the Influencer on Instagram. The requests GET methods scraping.get(“edge_media_to_caption”,None), edge_media_to_caption.get(“edges”, None), edge.get(“node”, None) and node.get(“text”) extract the value of the attributes “edge_media_to_caption”, “edges”, “node” and finally the description of photos or videos posted by the Influencer in the attribute “text”. The dumps JSON method json.dumps(edge_media_to_caption) decodes the value and saves the value of “text” in the variable edge_media_to_caption, as shown in Figure 9.

To navigate the edges, the function FOR for edge in edges accesses the “node” for each iteration of edge up to grab “text”.

Instagram Scraping for Edge_media_to_comment: The function IF if keys =  = “edge_media_to_comment”, a variable edge_media_to_comment stores the number of comments per post by the User of photos or videos posted by the Influencer on Instagram. The requests GET methods scraping.get(“edge_media_to_comment”, None) and edge_media_to_comment.get(“count”) extract the value of the attributes “edge_media_to_comment” and “count”, and save the value of “count”, i.e. the number of comments per post, in the variable edge_media_to_comment, as shown in Figure 10.

Instagram Scraping for Edge_liked_by: The function IF if keys =  = “edge_liked_by”, a variable edge_liked_by stores the number of likes per post by the User of photos or videos posted by the Influencer on Instagram. The requests GET methods scraping.get(“edge_liked_by”, None) and edge_liked_by.get(“count”) extract the value of the attributes “edge_liked_by” and “count”, and save the value of “count”, i.e. the number of likes per post in the variable edge_liked_by, as shown in Figure 11.

• Stage 5: Creating, writing and saving the Profile and Instagram Data

In this stage, once the profile data id, owner, edge_media_to_caption, edge_media_to_comment and edge_liked_by of the Instagram Influencer Target Group (see Table 2) and data extraction of graphql, user, edge_owner_to_timeline_media, edges and node ends, the Instagram data is created, written and stored in a CSV file, called “Instagram_data_not_clean.csv”.

The Python code Instagram Data Web Scraping (see Figure 12) starts with import json, import re and import pandas as pd.

To see the number of data, the number of columns, the data type of each column and the number of nulls of the CSV file, a number of PANDAS functions are necessary, as shown in Figure 12.

The function read_csv(“Instagram_data_not_clean.csv”) loads the data file in a variable df = pd.read_csv(“Instagram_data_not_clean.csv”).

The function DataFrame(df) transforms df into a Data Frame and loads it in the variable df = pd.DataFrame(df).

The function df.shape gets the number of rows (registers) and the number of columns (attributes) of the Data Frame.

The function df.info gets the types of data.

The function pd.isnull(df).sum( )) gets the total of null data per attribute.

The results of PANDAS functions: Shape, Info and isNull may be observed in Table 3a and 3b.

No missing data in the “Instagram_data_not_clean.csv” file, but the column Edge_media_to_caption has emojis, extra inverted commas, accents, special characters, uppercase and lowercase. Once all of this is eliminated, the identification is made manually, and the whole column is transformed into lowercase. An Excel file is created called “data_clean.xls”.

Therefore, the Instagram profile is saved in a CSV file, not clean, and then cleaned in an Excel file. As shown after the Shape, Info and isNull PANDAS results in Figure 12.

The function open(“Instagram_data_not_clean.csv”, “w”, newline = “) receives the following parameters: a file name, a letter “w” for a file writing and a new line parameter newline = ”', to add an empty line, as shown in Figure 13.

The WRITEROW [16, 17] function writes items in a sequence, separating them by a comma character. After creating the chain, the writerow( ) function writes “six columns” on the CSV file: “id”, “owner”, “edge_media_to_caption”, “edge_media_to_comment”, “edge_liked_by” and “isILandNSWused”, for the Publication ID, the Instagram Influencer name, the description of the photo or video posted by the Influencer on Instagram, the number of User comments per post, the number of User likes per post and a numerical attribute isILandNSWused, which is “1” if the publication belonged to the IL and NSW contain in the BoW and “0” if not, for each Influencer of the Target Group (see Table 2), respectively, as shown in Figure 13.

• Stage 6: Analyzing the GFL by the Influencer Target Group

In this stage, once the Instagram profile of the Influencer Target Group (see Table 2) is clean and loaded in an XLS file, as well as the BoW that contains the IL and NSW. The two of them, “data_clean.xls” and “BoW.xls”, are used as parameters in a function read_excel(“data_clean.xls”) and read_excel(“BoW.xls”) that loads the Excel data files in a variable file_clean = pd.read_excel(“data_clean.xls”) and file_BoW = pd.read_excel(“BoW.xls”). And then, it transforms into a Data Frame: DataFrame(file_clean) and DataFrame(file_BoW) that are loaded in the variables df_file_clean = pd.DataFrame(file_clean) and df_file_BoW = pd.DataFrame(file_BoW), respectively, as shown in Figure 13.

Thus, the XLS files contain the clean and lowercase text of the publications per post of the Influencer Target Group “data_clean.xls” (i.e. the Data Frame df_file_clean) and the IL and NSW, the ILandNSW language “BoW.xls” (i.e. the Data Frame df_file_BoW).

A numerical variable called isILandNSWused is defined, which initializes from zero.

The function FOR for i in df_file_clean.index swipes the Data Frame df_file_clean by its index to enter the column “edge_media_to_caption” for each row iteration of i. The value is stored in the variable df_caption_post.

The function IF if nan verifies the result of the function isCaptionPostNull(df_caption_post). If the result is “TRUE”, the post contains a “NaN'value” and therefore is ignored since the function re.search( ) cannot process these values. If the result is “FALSE”, the function check_ILandNSW(df_file_BoW,df_caption_post) is called, through which a value from one to zero is obtained if this is coincident or not with a BoW in the post. Therefore, the value of the function check_ILandNSW(df_file_BoW,df_caption_post) is assigned to the variable isILandNSWused; otherwise, the value of isILandNSWused remains as 0.

Then, with all the results, the row analyzed is saved through the function writer.writerow( ), as shown in Figure 13.

The function def check_ILandNSW(BoW, caption_post) has two parameters BoW and caption_post. The first represents the words belonging to the BoW. The second post was extracted from the Influencer post.

A numerical variable called isBoWinCaptionPost is defined, which initializes from zero. This variable has the function of defining whether there are ILandNSW words from the Data Frame df_file_BoW in the text of the column “edge_media_to_caption” of the Data Frame df_file_clean.

Then, a FOR for ILandNSW_word in df_file_BoW[“ILandNSW_words”] sweeps the column “ILandNSW_words” of df_file_BoW[“ILandNSW_words”] and extracts each IL and NSW “ILandNSW_word” of them. The function re.search(ILandNSW_word, caption_text) verifies whether that ILandNSW_word is contained or not in caption_text. If ILandNSW_word is contained, a value of one is assigned to the variable isBoWinCaptionPost; otherwise, the value remains zero. Finally, this function returns the value of the variable isBoWinCaptionPost (0 or 1), as seen in Figure 13.

The function def isCaptionPostNull(text)function verifies whether the text is or not “NaN”. As the “NaN” values are not the same as any other value, not even to the same “NaN”, one value returns a comparison with itself. If it is equal to itself, it returns a “TRUE”, and if not, a “FALSE”.

• Stage 7: Visualizing the Dashboard generated on Power BI

When the previous step ends and the data analysis is complete, different visualizations are conducted using the tool Microsoft Power BI. These visualizations seek to respond to the different management indicators and observe how data behave. Some of the results of this step are shown below.

The number of publications by language type. To calculate this indicator, the option “pie chart” is used, in which publications that comply or do not with the ILandNSW language are differentiated by percentage. “Inclusive” language type means that the language belongs to the ILandNSW language, while “non-inclusive” means the contrary, as shown in Figure 14.

The number of publications by Influencers and language type. The option “grouped bar chart” is employed to calculate this indicator, which shows two columns per Influencer. These columns show the number of publications by language type. The red column represents the publications that do not comply with the ILandNSW language, and the blue is those that comply with it. See Figure 15.

The number of likes. A visualization called “Tarjeta” (card) shows the total number of likes in all publications employed, as seen in Figure 16.

The number of comments. A visualization called “Tarjeta” (card) shows the total number of comments in all publications employed, as seen in Figure 17.

The number of publications. A visualization called “Tarjeta” (card) shows the total number of publications employed, as seen in Figure 18.

Publications with more likes. To calculate this indicator, the option “Table” is employed, which consists of a list with a description written by Influencers on each post, the user who wrote the publication and the number of likes of the same. This list is arranged in decreasing order by the number of likes, as shown in Figure 19.

Publication with more comments. To calculate this indicator, the option “Table” is employed, which consists of a list with a description written by Influencers on each post, the user who wrote it and the number of likes of the publication. This list is arranged in decreasing order by the number of comments, as shown in Figure 20.

The number of likes and comments made by Influencers. To calculate this indicator, the option “grouped bar and line chart” is used to show the relationship between the number of likes and the number of comments by an Influencer. See Figure 21.

Finally, the dashboard is generated after creating the charts, which aims to provide a global overview of all management indicators analyzed in the study. Figure 22 shows that the dashboard created allows for visualizing the results in such a way that they are more interactive or intuitive and for visualizing the diverse relationships that may exist between data; for example, between the number of likes and comments.

The dashboard contains two panels that allow for implementing filters to constrain data. If a field of these filters is selected, the data supplied to the other charts are automatically constrained to comply with the selected filter. At a glance, the “pie chart” shows a clear difference between the number of publications that meet the ILandNSW language, with only 7.5% of total publications complying with the ILandNSW language under study and 92.5% not complying. Thus, it is concluded that this type of language is not usually employed by Influencers in their publications on the SN Instagram.

Only four of the ten influencers occasionally use the ILandNSW language under study in their publications. Of these, the ones who use this ILandNSW language the most are Camila Recabarren (@camilarecabarrenoficial) and Raquel Calderón (@k3lcalderon), each of them with three posts, followed by Ignacia Antonia (@ignaciaa_antonia) with two and finally Joaquín Méndez with only one. In addition, of these four influencers, Camila Recabarren declares herself to be a feminist who supports the LGBT movement, but this is not reflected in her publications since more than 50% of her posts do not use IL and NSW.

The top places in the likes ranking are occupied by the posts of Fernanda Villalobos (@iamferv), Ignacia Antonia (@ignaciaa_antonia) and Paloma Castillo (@palomamami). Here, it may be observed that in the top ten rankings, there is no publication using the ILandNSW language under study. Additionally, a clear relationship between the number of likes and the types of language used can be observed: publications without the IL and NSW receive the most likes from followers. If the filter “type of language” is applied. In that case, the top publications for the type of language (belonging or not to the ILandNSW language under study) can be obtained, in which the difference between the number of likes among publications can be observed.

The number one post in the publication’s ranking does not belong to the ILandNSW language studied and was written by Fernanda Villalobos (@iamferv), having 1,765,918 likes. In turn, the top publication that contains this language is made by Ignacia Antonia (@ignaciaa_antonia), which has 192,195 likes. Thus, it is clear that there is a significant difference in the number of likes between posts not including and including the ILandNSW language. Meanwhile, it is observed in Figure 23 that the publications with ILandNSW language made by Ignacia Antonia (@ignaciaa_antonia), who has the post belonging to the ILandNSW language with the most likes, have more likes than those without the ILandNSW language studied.

A relationship is also found between the number of comments and the Type of Language: publications with more comments are those not belonging to the ILandNSW language. This can be observed in a ranking similar to the number of likes. As in the previous ranking, if the number of publications belonging to the ILandNSW language under study is analyzed, there is one publication by Camila Recabarren (@camilarecabarrenoficial) with 4,356 comments, while if looking at the ranking of publications not belonging to the ILandNSW language, there is one post by Paloma Castillo (@palomamami) that has 44,514 comments. Therefore, a significant difference in the number of comments between the top post of the different types of language may be observed, as shown in Figure 24.

In sum, more publications must comply with the ILandNSW language, and these publications also have more reactions from Users. Therefore, it is inferred that the use of IL and NSW still needs to be more widespread in the community of the SN Instagram since its use is not constant but only occasional. In addition, the posts (publications) that use the ILandNSW language under study receive little reactions from Users.

5. Conclusions

Regarding the Instagram influencers, after the identification and selection stages, only ten influencers from the top Influencers in Chile were identified, which is the critical point in this study. In total, data from 120 publications were extracted: Instagram Influencer name, publication ID, publication description (what the Instagram Influencer wrote on the post), the number of comments and the number of likes per post (publications) were obtained. Then, a numerical attribute isILandNSWused was added, which was “1” if the publication belonged to the IL and NSW and “0” if not.

Concerning the BoW, this was created before the text analysis stage and included words belonging to the Influencer Target Group ILandNSW language. This BoW may grow if new words used in the ILandNSW language studied are discovered. As for language, the BoW is contrasted with the content written by the Influencers to analyze if its words are contained in it.

Regarding the research questions of this study:

What is the level of use of IL among influencers in Chile on Instagram?

With the GFL Clustering Methodology implemented and the use of tools for processing the obtained data (Stages 1, 2, 3, 4 and 5), only 9 out of 120 publications were found to contain the ILandNSW language of the Influencer Target Group. After visualizing the information (Stages 6 and 7), it may be observed that of 120 posts made by Influencers on the SN Instagram, only 7.5% incorporated the IL and NSW in the descriptions written by diverse Influencers.

What words allow for determining the minimum degree of use of GFL? What is the relationship between Influencers and the use of ILandNSW language in digital marketing?

According to the results of this study, this type of ILandNSW language needs to be integrated into the various actions posted commonly on SN Instagram.

It is also observed that the publications with the most likes are comments that do not contain IL and NSW. Therefore, publications with this type of language could attract Chile’s Instagram influencers more.

In future work, the database (128 ILandNSW words) of the BoW of IL and NSW needs to be expanded to deepen the identification of the studied language since currently there are only guidelines for using this ILandNSW language rather than a list of words belonging to it. Creating an ILandNSW dictionary in other languages is also necessary for the future, as well as working with a larger pool of Influencers. Additionally, a web tool should be created to allow organizations to assess the use levels of ILandNSW language.

6. Managerial implications

The originality of this work lies in its attractive implications for nonprofit and for-profit organizations, government bodies and private enterprises in the measurement of the success of campaigns with an IM communicational strategy and to incorporate inclusive and non-sexist content for their consumers to contribute to society. Additionally, this study contributes guidelines for DC creators who aim to conduct strategic management with a market orientation, as well as for organizations who desire both to contribute to society through strategies for business communication and annual report and to define their online communication, product commercialization and branding strategies to generate a positive attitude toward the brand. Data could be retrieved from influencers as they are of free access regardless of the people involved.

7. Limitations and potential future research directions

Due to the limitations of the API of the SN Instagram, a limited number of text data were used, which allowed for retrieving the last 12 publications of each studied profile. In addition, this study only includes the Spanish language and is applied to a sample of influencers from Chile.

Another matter that could be addressed by further work is the implementation of an algorithm in Machine Learning to identify and classify ILandNSW language in such a way that complete sentences are retrieved, and their meaning is analyzed to know if what Influencers write genuinely belongs to the ILandNSW language under study.