SlideShare a Scribd company logo
1 of 26
Download to read offline
When E-commerce Meets Social Media:
Identifying Business on WeChat Moment
Using Bilateral-Attention LSTM
Tianlang Chen1, Yuxiao Chen1, Han Guo2, Jiebo Luo1
1. University of Rochester
2. Institute of Computing Technology, Chinese Academy of Sciences
E-commerce and Online Shopping
The Largest 10 Companies by Market Cap
WeChat Business/Non-business Moment
Examples of WeChat Moment. A red dotted box indicates a WeChat
Business Moment, a blue dotted box indicates a WeChat non-business
Moment.
1
幻灯片 4
1 >WeChat, one of the most well-known messaging platforms with 806 millions monthly active users 1 in China
>WeChat Business represents the activity for sellers to advertise their products, companies platforms or other services via WeChat.
>they can adveritise them by posting a WeChat Moment, which contains up to 9 images and a text, similar to Instragram post
, 2018/4/20
Task and Motivation
• Research goal:
 Identifying WeChat Business Moment
• Motivation:
 For online shopping lovers/addicts: push notification
service from these Moments to provide instant and
comprehensive information for potential purchases
 For online shopping haters: block these Moments because
they are nuisance
 A robust classifier is beneficial to both kinds of users.
 Task:
 Binary Classification
Overview
Overview of the proposed work
2
幻灯片 6
2 In particular, different from previous schemes that equally consider visual and textual modalities for a joint visual-textual
classification task, we start our work with a text classification task based on an LSTM network, then we incorporate a
bilateral-attention mechanism that can automatically learn two kinds of explicit attention weights for each word, namely
1) a global weight that is insensitive to the images in the same Moment with the word, and 2) a local weight that is sensitive to the
images in the same Moment. In this process, we utilize visual
information as a guidance to figure out the local weight of a word in a specific Moment.
>local weight represents semantic similarity between word and image environments, indicating the importance of word in a
Moment.
>global weight indicates the importance of word in the classification task.
, 2018/4/21
Architecture of Bilateral-Attention LSTM
Blue part:
 basic LSTM-based text
classification model
Green part:
 self-learned attention sub-network
 computes word’s global weight
Red part:
 image-guide attention sub-network
 computes word’s local weight
3
幻灯片 7
3 >The blue part is a combination of word embedding, LSTM and fully-connected layers for text classification.
>Compute the local and global weight and feed them into LSTM equation to guide the model predict
, 2018/4/21
Word Local Weight
 Meaning:
 semantic similarity between word and
image environment, representing the
importance of word in a Moment
 Unsupervised Image classification:
 ResNet-50 + K-means
 Image environment feature:
 feature construction based on
category
 Category-based Word feature:
 Bayesian model:
 Local weight:
Category name and illustration
(a) Flower (b) Meal (c) Cosmetics
(d) Chat screenshot
4
幻灯片 8
4 >Image environment feature and category-based word feature of a Moment are constructed based on image category. Therefore,
first do unsupervised image classification.
>Image environment feature = category distribution feature, is a n-dimensional feature where n represents the category number, for
each category, if there exist at least one image in this category, then it is set as 1.
>Category-based word feature (CC word feature) is a n-dimensional feature vector that records the correlation coefficient thetaWC
between the word and each specific image category, it is denoted as thetaW
>compute local weight based on Image environment feature and category-based word feature
, 2018/4/21
Word Global Weight
 Meaning:
 Representing the importance of a word in the classification task, unrelated to
the corresponding images of the Moment
 Example:
 “mountain” and “face mask” hold high global weight, “mountain” has strong
negative correlation with WeChat Business Moment, “face mask” has strong
positive correlation with WeChat Business Moment
 Word feature:
 Embedding feature extracted from word embedding layer
 Global weight
5
幻灯片 9
5 Global weight is self-learned from word embedding feature
, 2018/4/20
Word Final Weight
 Meaning:
 Representing the overall importance of a word
 Representation:

 Feeding:

6
幻灯片 10
6 The final weight is the multiplication of global weight and local weight, which achieves best performance
, 2018/4/20
Experiments
 Dataset:
 Totally 570 users, 37,359 WeChat Moments and 109,545 Moment images
 Labeled 10078 Moments, 4309 of them are WeChat Business Moments
 Randomly select 80% Moments as training/validation samples, 20% as test
samples
Vertical Experiment
Parallel Experiment
B-LSTM over other
textual models
B-LSTM over other
fusion schemes
7
幻灯片 11
7 >First demonstrates that LSTM achieves better performance than other text classification models (decision tree model , and another
two models that using LDA and doc2vec to extract feature)
>For the parallel experiment, we compare the model’s performance with late fusion, factorization machines, LSTM Q and CCR. For
late fusion, we fuse text feature and image environment feature (Category Distribution Feature) at the top of the model. For CCR,
we add a new loss term that imposes consistent constraints across visual and textual modalities. For factorization machines, we
replace the top fully-connected layers with factorization machines layers. For LSTM Q, we normalize the visual feature (CD feature),
and also implement common space mapping and element-wise multiplication.
>CD + LF = late fusion with category distribution feature (image environment feature)
, 2018/4/21
Visualization
Correlation between WeChat Business and each image category
8
幻灯片 12
8 Train a logistic regression model for image environment feature (category distribution feature), get the model’s learning weight of
each category to represent the correlation between category and WeChat Business
, 2018/4/21
Visualization
Typical words with their local weights in a Moment that contains
images of a single category (Sigmoid unit is replaced with ReLU
for a better display of words’ differences).
Visualization
Visualization of the bilateral attention mechanism
9
幻灯片 14
9 Two moments with same text and different images. Therefore, each word has same global weight but different local weight. Moment
A contains images related to cosmetic, it gives the last several words high local weights. Moment B contains images related to
cosmetic, it gives the first several words high local weights. In the end, the model predict Moment A as WeChat Business Moment, B
as WeChat non-business Moment
, 2018/4/20
Visualization
Visualization of the bilateral attention mechanism
10
幻灯片 15
10 Two Moments with same image and different texts. The model will in the end capture the most significant words and improve the
classification accuracy.
(Not very typical, this page can be dropped off if it is not necessary)
, 2018/4/20
Thanks!
Disclaimer: The dataset used in this study was not provided by Tencent.

More Related Content

Similar to When E-commerce Meets Social Media: Identifying Business on WeChat Moment Using Bilateral-Attention LSTM

INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKINGINTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
IJwest
 

Similar to When E-commerce Meets Social Media: Identifying Business on WeChat Moment Using Bilateral-Attention LSTM (20)

A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image AnnotationA Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
 
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
 
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
 
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image AnnotationA Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
 
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
A Multi Criteria Decision Making Based Approach for Semantic Image Annotation
 
A LINK-BASED APPROACH TO ENTITY RESOLUTION IN SOCIAL NETWORKS
A LINK-BASED APPROACH TO ENTITY RESOLUTION IN SOCIAL NETWORKSA LINK-BASED APPROACH TO ENTITY RESOLUTION IN SOCIAL NETWORKS
A LINK-BASED APPROACH TO ENTITY RESOLUTION IN SOCIAL NETWORKS
 
Seven Quality management & control tools
Seven Quality management & control toolsSeven Quality management & control tools
Seven Quality management & control tools
 
Multimedia And Contiguity Principles Casey Susan
Multimedia And Contiguity Principles Casey SusanMultimedia And Contiguity Principles Casey Susan
Multimedia And Contiguity Principles Casey Susan
 
Developing Movie Recommendation System
Developing Movie Recommendation SystemDeveloping Movie Recommendation System
Developing Movie Recommendation System
 
leewayhertz.com-Visual ChatGPT Conversational AI.pdf
leewayhertz.com-Visual ChatGPT  Conversational AI.pdfleewayhertz.com-Visual ChatGPT  Conversational AI.pdf
leewayhertz.com-Visual ChatGPT Conversational AI.pdf
 
Marvin_Capstone
Marvin_CapstoneMarvin_Capstone
Marvin_Capstone
 
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGESNEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
 
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGESNEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
 
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGESNEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
 
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGESNEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
 
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGESNEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
 
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGESNEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
 
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGESNEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
NEW ONTOLOGY RETRIEVAL IMAGE METHOD IN 5K COREL IMAGES
 
INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
INTELLIGENT SOCIAL NETWORKS MODEL BASED  ON SEMANTIC TAG RANKINGINTELLIGENT SOCIAL NETWORKS MODEL BASED  ON SEMANTIC TAG RANKING
INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
 
INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKINGINTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
INTELLIGENT SOCIAL NETWORKS MODEL BASED ON SEMANTIC TAG RANKING
 

More from Goergen Institute for Data Science

More from Goergen Institute for Data Science (14)

Vision and Language: Past, Present and Future
Vision and Language: Past, Present and FutureVision and Language: Past, Present and Future
Vision and Language: Past, Present and Future
 
Learning with Unpaired Data
Learning with Unpaired DataLearning with Unpaired Data
Learning with Unpaired Data
 
How to properly review AI papers?
How to properly review AI papers?How to properly review AI papers?
How to properly review AI papers?
 
Video + Language 2019
Video + Language 2019Video + Language 2019
Video + Language 2019
 
Video + Language
Video + LanguageVideo + Language
Video + Language
 
Computer Vision++: Where Do We Go from Here?
Computer Vision++: Where Do We Go from Here?Computer Vision++: Where Do We Go from Here?
Computer Vision++: Where Do We Go from Here?
 
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
A Selfie is Worth a Thousand Words: Mining Personal Patterns behind User Self...
 
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
When Fashion Meets Big Data: Discriminative Mining of Best Selling Clothing F...
 
Big Data Better Life
Big Data Better LifeBig Data Better Life
Big Data Better Life
 
Social Multimedia as Sensors
Social Multimedia as SensorsSocial Multimedia as Sensors
Social Multimedia as Sensors
 
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
Forever Young: A Tribute to the Grandmaster through a recount of Personal Jou...
 
Video+Language: From Classification to Description
Video+Language: From Classification to DescriptionVideo+Language: From Classification to Description
Video+Language: From Classification to Description
 
Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?
 
Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?Video + Language: Where Does Domain Knowledge Fit in?
Video + Language: Where Does Domain Knowledge Fit in?
 

Recently uploaded

Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
amitlee9823
 
FESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfFESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdf
MarinCaroMartnezBerg
 
Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
ZurliaSoop
 
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service BangaloreCall Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
amitlee9823
 
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
amitlee9823
 
Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...
Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...
Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...
amitlee9823
 
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night StandCall Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
amitlee9823
 
Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...
Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...
Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...
amitlee9823
 
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
amitlee9823
 

Recently uploaded (20)

Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
 
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
 
FESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdfFESE Capital Markets Fact Sheet 2024 Q1.pdf
FESE Capital Markets Fact Sheet 2024 Q1.pdf
 
Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Surabaya ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
 
Edukaciniai dropshipping via API with DroFx
Edukaciniai dropshipping via API with DroFxEdukaciniai dropshipping via API with DroFx
Edukaciniai dropshipping via API with DroFx
 
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service BangaloreCall Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
Call Girls Begur Just Call 👗 7737669865 👗 Top Class Call Girl Service Bangalore
 
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
Call Girls Hsr Layout Just Call 👗 7737669865 👗 Top Class Call Girl Service Ba...
 
Carero dropshipping via API with DroFx.pptx
Carero dropshipping via API with DroFx.pptxCarero dropshipping via API with DroFx.pptx
Carero dropshipping via API with DroFx.pptx
 
Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...
Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...
Call Girls Indiranagar Just Call 👗 7737669865 👗 Top Class Call Girl Service B...
 
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night StandCall Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
Call Girls In Bellandur ☎ 7737669865 🥵 Book Your One night Stand
 
CebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxCebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptx
 
Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...
Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...
Escorts Service Kumaraswamy Layout ☎ 7737669865☎ Book Your One night Stand (B...
 
Ravak dropshipping via API with DroFx.pptx
Ravak dropshipping via API with DroFx.pptxRavak dropshipping via API with DroFx.pptx
Ravak dropshipping via API with DroFx.pptx
 
Discover Why Less is More in B2B Research
Discover Why Less is More in B2B ResearchDiscover Why Less is More in B2B Research
Discover Why Less is More in B2B Research
 
(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7
(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7
(NEHA) Call Girls Katra Call Now 8617697112 Katra Escorts 24x7
 
Predicting Loan Approval: A Data Science Project
Predicting Loan Approval: A Data Science ProjectPredicting Loan Approval: A Data Science Project
Predicting Loan Approval: A Data Science Project
 
Mature dropshipping via API with DroFx.pptx
Mature dropshipping via API with DroFx.pptxMature dropshipping via API with DroFx.pptx
Mature dropshipping via API with DroFx.pptx
 
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
Digital Advertising Lecture for Advanced Digital & Social Media Strategy at U...
 
ALSO dropshipping via API with DroFx.pptx
ALSO dropshipping via API with DroFx.pptxALSO dropshipping via API with DroFx.pptx
ALSO dropshipping via API with DroFx.pptx
 
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
Junnasandra Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore...
 

When E-commerce Meets Social Media: Identifying Business on WeChat Moment Using Bilateral-Attention LSTM

  • 1. When E-commerce Meets Social Media: Identifying Business on WeChat Moment Using Bilateral-Attention LSTM Tianlang Chen1, Yuxiao Chen1, Han Guo2, Jiebo Luo1 1. University of Rochester 2. Institute of Computing Technology, Chinese Academy of Sciences
  • 3. The Largest 10 Companies by Market Cap
  • 4. WeChat Business/Non-business Moment Examples of WeChat Moment. A red dotted box indicates a WeChat Business Moment, a blue dotted box indicates a WeChat non-business Moment. 1
  • 5. 幻灯片 4 1 >WeChat, one of the most well-known messaging platforms with 806 millions monthly active users 1 in China >WeChat Business represents the activity for sellers to advertise their products, companies platforms or other services via WeChat. >they can adveritise them by posting a WeChat Moment, which contains up to 9 images and a text, similar to Instragram post , 2018/4/20
  • 6. Task and Motivation • Research goal:  Identifying WeChat Business Moment • Motivation:  For online shopping lovers/addicts: push notification service from these Moments to provide instant and comprehensive information for potential purchases  For online shopping haters: block these Moments because they are nuisance  A robust classifier is beneficial to both kinds of users.  Task:  Binary Classification
  • 7. Overview Overview of the proposed work 2
  • 8. 幻灯片 6 2 In particular, different from previous schemes that equally consider visual and textual modalities for a joint visual-textual classification task, we start our work with a text classification task based on an LSTM network, then we incorporate a bilateral-attention mechanism that can automatically learn two kinds of explicit attention weights for each word, namely 1) a global weight that is insensitive to the images in the same Moment with the word, and 2) a local weight that is sensitive to the images in the same Moment. In this process, we utilize visual information as a guidance to figure out the local weight of a word in a specific Moment. >local weight represents semantic similarity between word and image environments, indicating the importance of word in a Moment. >global weight indicates the importance of word in the classification task. , 2018/4/21
  • 9. Architecture of Bilateral-Attention LSTM Blue part:  basic LSTM-based text classification model Green part:  self-learned attention sub-network  computes word’s global weight Red part:  image-guide attention sub-network  computes word’s local weight 3
  • 10. 幻灯片 7 3 >The blue part is a combination of word embedding, LSTM and fully-connected layers for text classification. >Compute the local and global weight and feed them into LSTM equation to guide the model predict , 2018/4/21
  • 11. Word Local Weight  Meaning:  semantic similarity between word and image environment, representing the importance of word in a Moment  Unsupervised Image classification:  ResNet-50 + K-means  Image environment feature:  feature construction based on category  Category-based Word feature:  Bayesian model:  Local weight: Category name and illustration (a) Flower (b) Meal (c) Cosmetics (d) Chat screenshot 4
  • 12. 幻灯片 8 4 >Image environment feature and category-based word feature of a Moment are constructed based on image category. Therefore, first do unsupervised image classification. >Image environment feature = category distribution feature, is a n-dimensional feature where n represents the category number, for each category, if there exist at least one image in this category, then it is set as 1. >Category-based word feature (CC word feature) is a n-dimensional feature vector that records the correlation coefficient thetaWC between the word and each specific image category, it is denoted as thetaW >compute local weight based on Image environment feature and category-based word feature , 2018/4/21
  • 13. Word Global Weight  Meaning:  Representing the importance of a word in the classification task, unrelated to the corresponding images of the Moment  Example:  “mountain” and “face mask” hold high global weight, “mountain” has strong negative correlation with WeChat Business Moment, “face mask” has strong positive correlation with WeChat Business Moment  Word feature:  Embedding feature extracted from word embedding layer  Global weight 5
  • 14. 幻灯片 9 5 Global weight is self-learned from word embedding feature , 2018/4/20
  • 15. Word Final Weight  Meaning:  Representing the overall importance of a word  Representation:   Feeding:  6
  • 16. 幻灯片 10 6 The final weight is the multiplication of global weight and local weight, which achieves best performance , 2018/4/20
  • 17. Experiments  Dataset:  Totally 570 users, 37,359 WeChat Moments and 109,545 Moment images  Labeled 10078 Moments, 4309 of them are WeChat Business Moments  Randomly select 80% Moments as training/validation samples, 20% as test samples Vertical Experiment Parallel Experiment B-LSTM over other textual models B-LSTM over other fusion schemes 7
  • 18. 幻灯片 11 7 >First demonstrates that LSTM achieves better performance than other text classification models (decision tree model , and another two models that using LDA and doc2vec to extract feature) >For the parallel experiment, we compare the model’s performance with late fusion, factorization machines, LSTM Q and CCR. For late fusion, we fuse text feature and image environment feature (Category Distribution Feature) at the top of the model. For CCR, we add a new loss term that imposes consistent constraints across visual and textual modalities. For factorization machines, we replace the top fully-connected layers with factorization machines layers. For LSTM Q, we normalize the visual feature (CD feature), and also implement common space mapping and element-wise multiplication. >CD + LF = late fusion with category distribution feature (image environment feature) , 2018/4/21
  • 19. Visualization Correlation between WeChat Business and each image category 8
  • 20. 幻灯片 12 8 Train a logistic regression model for image environment feature (category distribution feature), get the model’s learning weight of each category to represent the correlation between category and WeChat Business , 2018/4/21
  • 21. Visualization Typical words with their local weights in a Moment that contains images of a single category (Sigmoid unit is replaced with ReLU for a better display of words’ differences).
  • 22. Visualization Visualization of the bilateral attention mechanism 9
  • 23. 幻灯片 14 9 Two moments with same text and different images. Therefore, each word has same global weight but different local weight. Moment A contains images related to cosmetic, it gives the last several words high local weights. Moment B contains images related to cosmetic, it gives the first several words high local weights. In the end, the model predict Moment A as WeChat Business Moment, B as WeChat non-business Moment , 2018/4/20
  • 24. Visualization Visualization of the bilateral attention mechanism 10
  • 25. 幻灯片 15 10 Two Moments with same image and different texts. The model will in the end capture the most significant words and improve the classification accuracy. (Not very typical, this page can be dropped off if it is not necessary) , 2018/4/20
  • 26. Thanks! Disclaimer: The dataset used in this study was not provided by Tencent.