Enhancing Invoice Number Retrieval in
Influencer Agency Transaction Records Using TF-IDF and SVC Model
Frisca Fitria
Universitas
President University, Indonesia
*Email: [email protected]
*Correspondence:
Frisca Fitria
|
DOI: 10.59141/comserva.v4i1.1339 |
ABSTRAK This research explore the need for efficient
invoice management in influencer marketing. The proposed solution using
TF-IDF and SVC Model to effectively locate and handle crucial transaction
information, including dates, brand names, invoice numbers, amounts,
frequency and category. By using this approach to significantly improve the
accuracy of identifying invoice numbers in influencer agency records. For the
result this method are evaluated using precision accuracy and F1 Score
metrics, to emphasize its effectiveness in current challenges of manual
extraction. Overall, this paper offers practical solution to enhanced
influencer agency transactions. Kata kunci: TF-IDF, SVC Model, Invoice
Number Retrieval, Influencer Agency, Transaction Records, Precision,
Accuracy. |
INTRODUCTION
Background
In modern advertising, influencer marketing has become
essential, signalling a shift in consumer-brand dynamics (Yeo et al., 2023).
The rising prevalence of influencer collaborations underscores the need for
efficient transaction record management, vital for financial accountability and
operational efficiency in influencer agencies. Given influencers' significant
impact on consumer preferences and purchasing decisions, effective management
of the endorsed products or brands is crucial. This research aims to enhance
this process through innovative techniques, such as TF-IDF and SVC Model,
addressing manual extraction challenges and offering practical solutions for
streamlined influencer agency transactions.
Problem
Statement
Manual extraction poses a hindrance to the seamless
financial workflows in influencer marketing. To address this efficiency gap,
this research advocates for the application of TF-IDF and SVC Model techniques.
These advanced methods aim to enhance the accuracy of identifying invoice
numbers within influencer agency records. This strategic approach aligns with
the industry's escalating reliance on influencers for effective brand
promotion, as highlighted by Yeo et al. in 2023. By leveraging innovative techniques,
we aim to streamline and optimize the process of managing financial
transactions within the dynamic landscape of influencer marketing.
Comparison
with existing approach
(Manjari, Rousha,
Sumanth, & Devi, 2020) revolutionize text summarization with TF-IDF and
Selenium, contrasting with our proposed system's focus on accuracy in
influencer agency transaction detail identification. While both utilize TF-IDF,
our approach is tailored to address challenges specific to influencer agency
transactions, optimizing accuracy in this domain.
(Yang & Long, 2023) enhance association rule mining with TF-IDF, whereas
our system utilizes TF-IDF and SVC Model for transaction detail identification
in influencer agency records. While both approaches utilize TF-IDF, our focus
on accuracy within influencer agency transactions addresses different
challenges and aims for specific optimizations.
(Pramono, Rohman, &
Hindersah, 2013) modify TF-IDF for topic extraction, contrasting with
our system's use of TF-IDF and SVC Model for transaction detail extraction in
influencer agency records. While both utilize TF-IDF, our approach addresses
challenges specific to influencer agency transactions, optimizing accuracy
within this domain.
(Kadhim, 2019) compares TF-IDF and BM25 for feature extraction,
whereas our system focuses on accuracy in influencer agency transaction detail
extraction using TF-IDF and SVC Model. While both approaches utilize TF-IDF,
our system addresses challenges specific to influencer agency transactions,
optimizing accuracy in this domain.
(Ngan, Lee, & Khor,
2023) automate paper assignment using TF-IDF and classification, while our
system enhances accuracy in influencer agency transaction detail extraction
with TF-IDF and SVC Model. Though both employ TF-IDF, our approach addresses
specific challenges in influencer agency transactions, optimizing accuracy for
this domain.
(Zheng Huang et al.,
2019) automate scanned receipt processing, whereas our proposed system aims
to enhance accuracy in identifying transaction details within influencer agency
records using TF-IDF and SVC Model. While both approaches focus on document
analysis, our system is tailored to address specific challenges in influencer
agency transactions, optimizing accuracy within this domain.
(Qiang & Zhong-min,
2022) extract symptom information from medical cases using TF-IDF and
Word2Vec, contrasting with our system's focus on accuracy in influencer agency
transaction detail extraction with TF-IDF and SVC Model. While both utilize
TF-IDF, our approach addresses challenges specific to influencer agency transactions,
optimizing accuracy within this domain.
(Nugawela, Abeywardena,
& Mahaadikara, 2022) extract tabular data from images algorithmically,
while our proposed system enhances accuracy in influencer agency transaction
detail extraction using TF-IDF and SVC Model. While both focus on information
extraction, our approach is tailored to address challenges specific to
influencer agency transactions, optimizing accuracy within this domain.
(Qisong et al., 2023) automate NOTAM information extraction using NLP,
whereas our system focuses on accuracy in influencer agency transaction detail
extraction using TF-IDF and SVC Model. While both aim for automation, our
approach addresses challenges specific to influencer agency transactions,
optimizing accuracy within this domain.
Alisa et al. (2021) extract immunosuppressive
properties from biomedical texts, contrasting with our system's focus on
accuracy in influencer agency transaction detail extraction using TF-IDF and
SVC Model. While both tackle information extraction, our approach is tailored
to address challenges specific to influencer agency transactions, optimizing
accuracy within this domain.
(Melnyk, Huymajer,
Huemer, & Galler, 2023) propose a digital documentation system for
construction management, while our system focuses on accuracy in identifying
transaction details within influencer agency records using TF-IDF and SVC
Model. While both aim to enhance efficiency, our approach addresses specific
challenges in influencer agency transactions, optimizing accuracy within this
domain.
(TongNan Huang, 2023) introduces blockchain-based transaction database
encryption, contrasting with our system's focus on accuracy in influencer
agency transaction detail extraction using TF-IDF and SVC Model. While both aim
to enhance security and reliability, our approach addresses challenges specific
to influencer agency transactions, optimizing accuracy within this domain.
(Li et al., 2020) propose a quaternary market transaction model for the
electric heating market, contrasting with our system's focus on accuracy in
influencer agency transaction detail extraction using TF-IDF and SVC Model.
While both propose innovative transaction models, our approach addresses
specific challenges in influencer agency transactions, optimizing accuracy
within this domain.
Ni et al. (2023) introduce FLUID for continuous
transaction processing in blockchains, while our system focuses on accuracy in
identifying transaction details within influencer agency records using TF-IDF
and SVC Model. While both aim to enhance efficiency in transactions, our
approach addresses challenges specific to influencer agency transactions,
optimizing accuracy within this domain.
(Reddy & Kumar,
2022) compare algorithms for credit card fraud detection, contrasting with
our system's focus on accuracy in influencer agency transaction detail
extraction using TF-IDF and SVC Model. While both aim to improve transaction
security, our approach addresses specific challenges in influencer agency
transactions, optimizing accuracy within this domain.
Overall, while existing research proposes innovative
approaches to various transaction-related tasks, this research specifically
targets accuracy in identifying transaction details within influencer agency
records, utilizing TF-IDF and SVC Model tailored to this domain's challenges.
Furthermore, this research use detailed evaluation
metrics, including accuracy, precision, F1 Score metrics and processing speed,
provide a comprehensive assessment of the proposed solution's effectiveness.
Research
aim and objectives
This research is dedicated to the rapid identification
of invoice numbers within influencer agency records. Through the application of
TF-IDF and SVC Model instead of Cosine Similarity, this approach prioritizes
accuracy and confronts challenges linked to manual extraction. The core
objective revolves around improving transaction record management, specifically
within the domain of influencer marketing, by presenting practical and
effective solutions. The incorporation of advanced techniques is geared towards
streamlining processes and offering an efficient means of retrieving invoice
numbers within the dynamic landscape of influencer agency transactions.
Limitations
This research utilizes TF-IDF and the SVC Model to
improve invoice number retrieval, acknowledging certain limitations. The study
predominantly evaluates the accuracy and precision of the algorithms with a
dataset of 20 samples, raising concerns about the generalizability of findings
to real-world scenarios. Variations in actual influencer marketing data may
impact outcomes. The evaluation emphasizes efficiency through mathematical
metrics, potentially overlooking practical challenges inherent in diverse
influencer marketing situations. These limitations underscore the importance of
a comprehensive analysis, recognizing the need for broader considerations
beyond the specific context of this research.
Hypothesis
Leveraging TF-IDF and the SVC Model for invoice number
retrieval is anticipated to significantly enhance accuracy and precision in
contrast to manual extraction. The proposed framework aims to achieve
heightened efficiency in managing influencer agency transaction records.
Outline
of the Research
This study introduces a practical solution aimed at
enhancing the identification of invoice numbers within influencer agency
transaction records. As the demand for influencer marketing continues to surge,
the need for an efficient invoice management process becomes imperative. By
leveraging TF-IDF and the SVC Model, this research addresses the crucial task
of accurately locating and extracting invoice numbers. Additionally, it
confronts challenges associated with manual extraction, evaluating the proposed
approach using precision and accuracy metrics. Through a comparative analysis
with existing methods in automated information extraction and invoice
management, this paper positions itself as a faster and more practical solution
for identifying invoice numbers in influencer agency records. The overarching
goal is to improve the efficiency and effectiveness of invoice management
processes.
METHOD
The methodology employed in this research is
meticulously designed to bolster the retrieval of invoice numbers in influencer
agency transaction records through the synergistic utilization of TF-IDF and
Support Vector Classification (SVC) Model techniques. The methodology has been
refined based on comprehensive insights from the research details provided
above. The outlined steps encompass data collection, pre-processing, TF-IDF
calculation, SVC Model implementation, and a thorough evaluation using diverse
metrics.
1)
Data
Collection
The initial step involves gathering a diverse dataset
of influencer agency transaction records from the internal database named
"Cyclone Management." The dataset encompasses essential details such
as including dates, brand names, invoice numbers, amounts, frequency and
category.
2) Pre-processing of Data
The collected transaction records undergo a meticulous
pre-processing phase. This involves cleaning and standardizing the records to
ensure consistency. The pertinent information, including dates, brand names,
invoice numbers, amounts, frequency and category, is extracted from the
dataset.
3)
TF-IDF
Calculation
a.
Term
Frequency
The TF for each term in the documents stored in the
"Cyclone Management" database is computed using the formula:
b.
Inverse
Document Frequency (IDF)
The IDF for each term across
all documents in the "Cyclone Management" database is calculated
using the formula.
c.
TF-IDF
Score
The
TF-IDF scores for each term in the sample documents are determined using the
formula:
4) SVC
Model Implementation
The
Support Vector Classification (SVC) Model is initialized and optimized using
grid search and cross-validation to achieve the best-performing model.
a. Linear
Kernel
The
linear kernel is defined by the dot product of the input samples.
b.
Radial Basis Function (RBF) Kernel
The RBF kernel, also known
as the Gaussian kernel, measures the similarity between two samples based on
the Euclidean distance.
c.
Decision Function Values
In
the specific context, the decision function values are calculated for the test
set (Xtest) using the trained SVC model (best_svc). These values serve as a
basis for predicting the class labels of the test samples. The grid search
utilized during the model training optimizes hyperparameters, such as C for
regularization, ensuring the SVC model effectively learns and generalizes from
the training data to make accurate predictions on new, unseen samples.
5) Predictions
and Ranking
a. Prediction
Function
The prediction function applies the trained SVM
model (best_svc) to the TF-IDF vectors of the testing data (Xtest). The SVM
model utilizes its decision function, which was learned during training, to
assign a predicted label to each document in the testing set. The predicted
labels are stored in the vector� y.
b. Ranking
Function
Ranked_indices=
The
ranking function operates on the decision function values produced by the
trained SVM model (best_svc) for the TF-IDF vectors of the testing data
(Xtest). These decision function values represent the model's confidence or
certainty about the classification of each document.
The
function uses the argsort operation to sort the indices of the testing samples
in ascending order based on their decision function values. The result is the
vector ranked_indices where each element corresponds to the position of a
document in the sorted order. The lower the index, the higher the decision
function value, indicating a higher confidence in the predicted category.
6) Evaluation
Metrics
a. Accuracy:
Accuracy
measures the proportion of correctly predicted instances among the total
instances. In the code, it is calculated by dividing the number of correct
predictions (accuracy_score) by the total number of predictions.
b.
Precision:
Precision evaluates the accuracy of positive
predictions. It is calculated by dividing the number of true pos.
c.
F1 Score:
F1 Score is the harmonic mean of precision and
recall. It considers both false positives and false negatives. The code
calculates it using the precision and recall scores. The formula emphasizes the
balance between precision and recall.
7) System
Implementation
a. Invoice
Number Retrieval
The
identified documents in the "Cyclone Management" database are
scanned, and the invoice numbers are extracted as potential matches for each
set of sample data.
b. Integration
into Transaction Record Management System:
The
TF-IDF and SVM retrieval system is seamlessly integrated into the existing
transaction record management system. This integration ensures compatibility
through the design of standardized data formats and interfaces, facilitating
smooth interaction with other modules of the system.
8) Processing
Speed Calculation
The
time taken for the entire process, encompassing both manual extraction and the
proposed system, is recorded. The speedup percentage, a measure of efficiency
improvement, is calculated using the following formula:
A
higher speedup percentage indicates a more efficient and faster system,
showcasing the advantages of the proposed TF-IDF and SVC Model-based approach.
RESULT
AND DISCUSSIONS
The
result and discussion of this research are presented values for the retrieval
of invoice numbers in influencer agency transaction records highlight several
key areas including evaluation metrics, brand analysis, dashboard visualization
and processing speed comparison.
1. �Evaluation
Metrics
Accuracy of TF IDF
on proposed methods is 91,15%, it is relatively provide high correct in
predicting influencer agency transaction. Precision stands at 100% it is
accurate and reduce the chances of false positive. F1 Score 0.0 the model is
struggles to achieve a balance between precision and recall.
2. Brand Analysis
It shows result after implement TF-IDF and SVC Model. The data shows
including rank, brand name, frequency rank and decision values rank. Consist of
2 table including top 10 brands with highest frequency and top 10 brands with
10 lowest frequency.
Each table includes of relevant information in providing insight to
follow up different brands within the transaction records.
3. Dashboard Visualization
This section
show visualization such as top 10 brands based on frequency and information to
enhance order frequency.
Information are
included discount for brand based on their frequency of transactions, discount
given has term and conditions such as low and highest get differ discounts only
active for 3 months, suggestions given based on conditions the available data on
influencer database. This dashboard has aims as data decision making for follow
up brand based on transaction records with more targeted.
4. Processing Speed Comparison
Invoice number still saved manually it needs to compare it one another
it spend 20 minutes, this proposed system only takes 0,44 minutes ~ 1 minutes.
Speedup Percentage: [(Time taken for Manual Extraction - Time taken for
Proposed System) / Time taken for Manual Extraction] * 100
Proposed methods are faster rather than manual extraction it is improve
97,81% compare with manual extraction.
CONCLUSION
The application of
TF-IDF and SVC Model for invoice number retrieval in influencer agency
transaction records has significant advancements in accuracy and efficiency.
With 91.15% accuracy and a precision score of 100%, it has good performance in
accurately identifying invoice numbers. While F1 Score has challenge in
achieving a balance between precision and recall. This research provide
dashboard for facilitate data visualization as solution to improves transaction
record management within the influencer marketing domain especially in follow
up brand for next collaboration based their data on invoice record. This
integration to enhances brand analysis and order frequency insights. This
solution only took 0,44 minutes, it already improved in processing speed which
before has 20 minutes it decrease time taken with 97.81% compared with manual
extraction. Future research may be maximize the model to achieve balance
between precision and recall to enhancing the overall performance. Also
exploring integration of Natural Language Processing (NLP) can be areas for
further improvements. Overall, this research presents a practical and effective
solution to enhance invoice management processes in influencer agency
transactions, contributing to rapidly and growing of financial workflows in
realm of influencer marketing.
BIBLIOGRAPHY
Huang, TongNan. (2023). Transaction Database
Encryption Technology based on Blockchain Technology. 2023 8th International
Conference on Information Systems Engineering (ICISE), 342�345. IEEE. doi: 10.1109/ICISE60366.2023.00078
Huang, Zheng, Chen, Kai, He, Jianhua, Bai, Xiang,
Karatzas, Dimosthenis, Lu, Shijian, & Jawahar, C. V. (2019). Icdar2019
competition on scanned receipt ocr and information extraction. 2019
International Conference on Document Analysis and Recognition (ICDAR),
1516�1520. IEEE. doi: 10.1109/ICDAR.2019.00244
Kadhim, Ammar Ismael. (2019). Term weighting for
feature extraction on Twitter: A comparison between BM25 and TF-IDF. 2019
International Conference on Advanced Science and Engineering (ICOASE),
124�128. IEEE. doi:
10.1109/ICOASE.2019.8723825.
Li, Shupeng, Liu, Yingchao, Ren, Shuai, Huo, Xianxu,
Yu, Jiancheng, Li, Qing, & Zhou, Ying. (2020). New Assumption of
Regenerative Electric Heating Market Model from the Perspective of Transaction
Cost. 2020 Asia Energy and Electrical Engineering Symposium (AEEES),
375�379. IEEE. doi:
10.1109/AEEES48850.2020.9121468.
Manjari, K. Usha, Rousha, Syed, Sumanth, Dasi, &
Devi, J. Sirisha. (2020). Extractive Text Summarization from Web pages using
Selenium and TF-IDF algorithm. 2020 4th International Conference on Trends
in Electronics and Informatics (ICOEI)(48184), 648�652. IEEE. doi: 10.1109/ICOEI48184.2020.9142938
Melnyk, Oleksandr, Huymajer, Marco, Huemer, Christian,
& Galler, Robert. (2023). Digitalization in the Construction Industry: The
Case of Documentation and Invoicing in Tunneling. 2023 IEEE 25th Conference
on Business Informatics (CBI), 1�10. IEEE.
doi: 10.1109/CBI58679.2023.10187588
Ngan, Seon Choon Han, Lee, Ming Jie, & Khor, Kok
Chin. (2023). Automating Conference Paper Assignment Using Classification
Algorithms Incorporated with TF-IDF Vectorisation. 2023 IEEE Symposium on
Industrial Electronics & Applications (ISIEA), 1�6. IEEE. doi: 10.1109/ISIEA58478.2023.10212219
Nugawela, Methmi, Abeywardena, Kavinga Yapa, &
Mahaadikara, Hansika. (2022). Algorithmically Navigating Complex Tabular
Structures in Images for Information Extraction. 2022 3rd International
Informatics and Software Engineering Conference (IISEC), 1�6. IEEE. doi: 10.1109/IISEC56263.2022.9998220
Pramono, Luthfan Hadi, Rohman, Arief Syaichu, &
Hindersah, Dan Hilwadi. (2013). Modified weighting method in TF* IDF algorithm
for extracting user topic based on email and social media in Integrated Digital
Assistant. 2013 Joint International Conference on Rural Information &
Communication Technology and Electric-Vehicle Technology (RICT & ICeV-T),
1�6. IEEE. doi:
10.1109/rICT-ICeVT.2013.6741547
Qiang, Cheng, & Zhong-min, Du. (2022). A NLP
Application of Automated Symptom Information Extraction from TCM Medical Cases.
2022 IEEE 2nd International Conference on Computer Systems (ICCS),
35�39. IEEE. doi:
10.1109/ICCS56273.2022.9988199.
Qisong, Hu, Zixuan, Wu, Wen, Yin, Minrui, Chen,
Guoqiang, Chen, & Yang, Yang. (2023). Research on NOTAM Information
Extraction of Civil Aviation with NLP. 2023 IEEE 5th International
Conference on Civil Aviation Safety and Information Technology (ICCASIT),
520�523. IEEE. doi:
10.1109/ICCASIT58768.2023.10351768
Reddy, P. Raghavendra, & Kumar, A. Sivanesh.
(2022). Credit Card Fraudulent Transactions Prediction Using Novel Sequential
Transactions by Comparing Light Gradient Booster Algorithm Over Isolation
Forest Algorithm. 2022 2nd International Conference on Innovative Practices
in Technology and Management (ICIPTM), 2, 563�567. IEEE. doi: 10.1109/ICIPTM54933.2022.9754211.
Yang, En, & Long, Zhaohua. (2023). Research on the
Weighting Method Based on Tf-IDF and Apriori Algorithm. 2023 IEEE 6th
International Conference on Information Systems and Computer Aided Education
(ICISCAE), 1003�1005. IEEE. doi:
10.1109/ICISCAE59047.2023.10393523.
|
|
|
