Which elements of the mobile banking app are critical from a customer perspective? What aspects receive the most and the least appreciation from customers? I explore customers’ reviews to understand their experience with mobile apps on a sample of European significant banks from 2010 to 2022. 

**Data**: customer reviews scraped from Google Play Store site in Python (google-play-scraper library).

**Time frame**: 2010-2022.

**Analysis**: preprocessing, review language detection (*langdetect*) and noun/adjectives detection (*spaCy*), word counts (*NLTK*).

**Visualization**: [Tableau](https://public.tableau.com/app/profile/arina.tsirkuleva/viz/Mobilebankingappreviews/mobile_app).

**Description**: I scraped customer reviews about the use of mobile bank applications. I
focused on European
significant banks that are directly supervised by the ECB ([list
of banks](https://www.bankingsupervision.europa.eu/ecb/pub/pdf/ssm.listofsupervisedentities202104.en.pdf?d62ded18a8c05df230e1a8132ead3a1b)).

For each bank, I found a link to the main
mobile bank app in Google Play Store. Then, with the ***Google Play
scraper*** library for Python, I
collected all customer
reviews for the period 2010-2022. The reviews are written in many languages,
hence, I detected the language
of each review with ***langdetect*** library in Python. After, I identified the most common language
of the reviews, and for the most part, one language dominates the reviews. However, for the banks of Greece, both English and Greek are almost equally important (see figure
below), so I decided to
keep them in the analysis.

I pre-processed the review in ***spaCy***
using available multilingual pre-trained [pipelines](https://spacy.io/models).
Due to the unavailability
of pipelines for several languages (in particular, Latvian, Finnish, and Estonian), I excluded these banks from the
analysis. The preprocessing stage included lowering text and removal of generic stopwords (spaCy's list
for each language) and some words
that merely describe the context and are not interesting for visualization
(such as, "bank"
and "application" in all languages involved in the analysis.

Next, with ***spaCy***, I
selected only nouns and
adjectives. Nouns describe the nature of problems and the foci of customers'
appreciation. Adjectives, instead allow grasping the emotional attitudes toward
the bank product, the mobile app. I aggregated the obtained lists of nouns/adjectives by customer rating, and with *NLTK Counter*
defined the most
frequent words (10 for nouns and 10 for adjectives) for each rating.

Finally, in ***Tableau***, I visualized the average score of the
mobile app over the period from 2010 to 2022\*, and the most frequent remarks
left by customers. The rating slider on the right side changes the distribution of remarks, allowing
this way to distinguish the most and the least critical aspects of apps'
functioning according to the
customers.

\*consider
that this might be different from the actual mobile app score because I analyzed only customer ratings with written reviews.


Customer experience with mobile banking apps

How performative is the language of the ECB? Through language, we not only describe the world around us but influence it as well (‘doing by saying’ in John Austin’s words). The best illustration of language performativity is the effect of the pledge to do “whatever it takes” by ECB President Mario Draghi in 2012. To understand the empowerment of the ECB across the years, I focus on the performativity of its communication from 1999 to 2021, namely on the modality of the executives' utterances.

The European Central Bank (ECB) has been devised as
a technocratic entity, yet, in the course of its 22-year history, the ECB takes
the lead in accelerating and driving European integration. Especially during
the financial crisis, the ECB is recognized to be the only body of the European
Union with the real “institutional ability” and power (Brunnermeier, James,
& Landau, 2016; Chang, 2015; Tortola, 2020).

The organizational and functional machinery of the ECB
has been initially outlined in the annex to the Maastricht Treaty,
the foundational document of the European Union. The entity had a clear
functional mandate. Yet, these "on paper" functions have been enacted over
years by the executive team of the
ECB. Members of the Executive Board represent the central bank and often
communicate in its name, shaping a certain organizational identity. To
understand the evolution and empowerment of the ECB I focus on the
modality of the executives' communication.

In linguistics, modality captures the speaker’s
attitude to reality or truth. A modal expression may convey that something is
desirable, necessary, or permissible. There are different classifications of
modalities, yet two are the most consistent:

*   **epistemic** **modality**
    that accounts for knowledge, beliefs, or credence in a proposition. This
    modality may be expressed as a possibility, necessity, and prediction.



*   **deontic** **modality**
    that describes how the world should be according to certain norms, and
    expectations. This modality can be expressed as a commitment to do something, advice, and obligation.

Hence, I explore how the ECB communication evolves
across these modalities.

**Data**: [ECB Speeches dataset](https://www.ecb.europa.eu/press/key/html/downloads.en.html).

**Time** **frame**:
1999-2021.

**Analysis**:
preprocessing (*NLTK, spaCY*), modality parsing (*spaCy*) - for this step I created
a dictionary that combines different ways of modality expression (e.g.,
possibility with modals *may*, *might*; in phrases "(We/ECB)
acknowledge(s) that..." etc.). It is an ongoing project, so I cannot publish the final dictionary yet.

**Preliminary** **results**:

![](/images/Epistemic.png)

![](/images/Deontic.png)

Performative language of the ECB

How central banks and central bankers are connected? I extract the most mentioned entities in the central banks’ communication from 2007 to 2021 to understand patterns of their interaction and coinfluence.

**Data**: Central
banks’ speeches from the
[BIS site](https://www.bis.org/cbspeeches/index.htm).  Scraped with Selenium (Python) in the end of 2021.

**Time frame**: 2007
– 2021.

**Analysis**: preprocessing (*spaCy*), named-entity recognition (*spaCy*), network creation (*NetworkX*), Louvain method for community detection.

**Visualization**: Pyvis ([organisations](https://arinatsr.github.io/cbs_org/)) and ([persons](https://arinatsr.github.io/cbs_pers/)) \[*suggested to use Force Atlas2 solver*].

**Description**: I build on a
corpus of central bankers’
speeches from 2007 to 2021.

In the analysis, I use ***spaCy*** to parse named entities (i.e.,
real-world objects, such as organizations, people, and geographic locations). In
particular, I focus on organizations and persons to understand which
entities/personalities central banks refer to the most in their communication. As a result of parsing, I get
a table that connects a central bank that delivers communication (source
column) with organizations/personalities that are mentioned in this particular
communication (target column). I exclude all self-references, automatically
(using [ECB
abbreviations glossary](https://www.ecb.europa.eu/services/glossary/html/glossa.en.html)) and manually resolve the
most frequent coreferences (i.e., different spellings of the same entity, such
as “us federal reserve”, “federal reserve system”, “fed”), and apply a
threshold of 10 word occurrences.

Using the source-target table, I generate two networks in ***NetworkX*** and visualize
them as interactive graphs with ***Pyvis***,
one for organizations and another for personalities.


Global network of central banks and central bankers

In a common monetary policy framework, how distant are European national central banks and the ECB, the supranational monetary authority? I explore the proximity of their communication in the period from 1999 to 2020.

**Data**: Introduction to annual reports of European central banks
from 1999 to 2020 (manually collected from officieal sites).

**Analysis**: Hellinger distance applied to topics defined
[previously](https://public.tableau.com/app/profile/arina.tsirkuleva/viz/CBsCommunication/CentralbanksCommunication).

**Visualization**: [Tableau.](https://public.tableau.com/app/profile/arina.tsirkuleva/viz/Centralbanksdistance/Distance_ECB)

**Description**: I build on a corpus of introductions to annual
reports of European central banks (CBs) that constituted the Eurosystem, the monetary authority of the Eurozone. Eurosystem comprises
the European Central Bank (ECB) and national central banks (NCBs) that adopted
the euro as a single currency. In
the analysis, I address the communication of 10 NCBs that launched the euro in 1999\*, and the CB of Greece that joined the Eurosystem in 2001.

I use [previously](https://public.tableau.com/app/profile/arina.tsirkuleva/viz/CBsCommunication/CentralbanksCommunication) defined 25 topics (main themes of a text collection) to
measure pairwise distances between the ECB and national central banks (NCBs). As a result of topic modeling, each
document of the text collection is quantified as a distribution of topics,
i.e., a list combining topic indices and topic shares in the document. Hence, I
apply Hellinger distance metrics on these distributions to measure distances between text pairs
(introductions to annual reports, in my case). Higher values correspond to more distance, and lower – to
more similarity.

I visualize the
results as an interactive dashboard in ***Tableau*** that shows the distance dynamics between individual NCBs and
the ECB.

\* The CB
of Portugal, the 11th
NCB to start the Eurosystem, is excluded from the analysis for the lack of data.


Distance between the ECB and national central banks

How does the communication of European central banks evolve? To what issues do they pay attention? All together and individually? I explore the building blocks of central banks' communication by applying topic modeling to the corpus of their annual report from 1999 to 2020.

**Data**: Introductions to annual reports of European central banks from 1999 to 2020 (collected manually from official sites).

**Analysis**: preprocessing (NLTK, spaCy), phrases detection (Gensim), topic modeling (Gensim).

**Visualzation**: [Tableau](https://public.tableau.com/app/profile/arina.tsirkuleva/viz/CBsCommunication/CentralbanksCommunication).

**Description**: I build on a corpus of introductions to
annual reports of European central banks (CBs) that constituted the Eurosystem, the monetary authority of the Eurozone.
Eurosystem comprises the European Central Bank (ECB) and national central banks
(NCBs) that adopted the euro as a single currency. In the analysis, I address the communication of 10 NCBs that launched the euro in 1999\*, and the CB of Greece that joined the Eurosystem in 2001.

I split texts into chunks of comparable size, first
dividing texts into paragraphs and then merging paragraphs that are too small.
After, I train the LDA topic model (***Gensim***
implementation). Topic modeling is an unsupervised clustering algorithm that defines the main themes of a text collection. I fit the model of 25 topics on the whole corpus, and these topics illustrate the focal points
of CBs' attention in the period from 1999 to 2020.

In ***Tableau***, I
visualize topics’ evolution
over years.
The visualization
allows to compare the use of topics by each CB with the annual average of all CBs (upper chart).
As well, I show how the attention
to each topic changes for
every CB (lower
chart).

\*The CB
of Portugal, the 11th
NCB to start the Eurosystem, is excluded from the analysis for the lack of data.


Projects

Customer experience with mobile banking apps

Performative language of the ECB

Global network of central banks and central bankers

Distance between the ECB and national central banks

Building blocks of central banks' communication