Meta background-color:: blue collapsed:: true
- [[Contents]]
- WIP2
- Surveys Sheet
- [Mac Folder](/Users/twenythree/Library/CloudStorage/GoogleDrive-ninjatxgaming@gmail.com/My Drive/CBS/Thesis)
- LEFT OFF #2024-04-11
  - ((660ae28c-e26a-4e02-a721-1c20ad247993)) id:: 660ae81b-3f51-41ac-aed0-4653d566c346
    - Left off: Page 8
  - ((65f354a5-9909-4d53-b940-a214a0ae479c))
- WIP Google Doc OLD.
- Tags
  - #personas
- [[4Airidas]]
- Other Links background-color:: purple collapsed:: true
For Elias (last updated: [[2024-03-14]]) background-color:: blue collapsed:: true
- {{embed ((662cca5d-9eff-4f0c-bb8d-bd9f649978db))}}
- Generally - my collected assets collapsed:: true
  - {{embed ((65ecc30b-520a-4735-9af1-f85986f164b8))}}
Code background-color:: purple collapsed:: true
- NOTE: Rosebud data
- NOTE: reddit scraping
- v2 simulation changes
  - v2
    - Local & Cloud
      - reversed
      - retrieval
        
        dynamic
        
        static
      - lower context (lower chunks in ctx)
  - v3
    - [[Chain of thought]]
    - rosebud data
    - temperature
WIP background-color:: purple collapsed:: true
- .
Selections background-color:: blue id:: 65a83ceb-a2e2-48ae-aacd-3634889b159f collapsed:: true
- About our Topic collapsed:: true
  - Aim to understand [[consumer preferences]].
  - As a tool for [[market research]]
    - Existing tools for market research id:: 65b0154b-de6a-4c32-92ce-47984c12c819
      - surveys
      - conjoint studies
      - focus groups
      - proprietary data sets
  - LLMs adopting personas to do market research
  - AI-Based Foundation Models as Alternates to Human Data Sources
- RQ collapsed:: true
  - How LLM's can be used to do market research?
  - Reprensentable digital personas
  - To what extent can LLM driven digital personas replace real people in market research?
    - To demonstrate the extent LLM driven digital personas can match real people in market research.
  - POPULAR USE CASES of [[conjoint analysis]] (src: foooter) id:: 65b69b6e-1907-40b4-92b8-422015faac08 collapsed:: true
    - 🔸 Market Research
    - 🔸 Text Analysis
    - Voice of Customer
    - Conjoint Analysis
    - Employee Tech Experience
    - Product Naming
    - Well-being at Work
    - Customer Experience
    - Management
    - Employee Experience Management
    - Exit Interviews
    - Brand Tracking
    - Pricing Research
    - NPS Software
    - Customer Surveys
- Unclear background-color:: purple collapsed:: true
  - Sample sizes collapsed:: true
    - {{embed ((65ecd8ff-014b-4cf1-91a3-91af84fd84d9))}}
    - From ((65ecc4f4-b5af-4f3b-b286-3b8787ec422e)) collapsed:: true
      - Each experimentally created subject was presented with each of the five scenarios: Neutral plus each of the four options offered as the status quo. Each call is a separate API call, with no “knowledge” of the other scenarios passed between calls. Because the agents are not interacting in any way, the experiment can proceed in parallel. Each of the 100 agents is presented in the five scenarios, creating 500 observations. Figure 4 shows the distribution of responses by framing.
  - [[Synthetic Data]]
- ◾ MR Methodology id:: 65c3f123-26dd-4fb9-899f-839c9376ce1a collapsed:: true
  - Quantitative background-color:: green collapsed:: true
    - ✔ [[Kano Model Mapping]]
    - Observational Studies
    - Surveys collapsed:: true
      - Todo
        
        ((65f1b7b9-a6f8-429b-87c9-fad9956047f6))
        
        ((65f1b8c6-3cf3-4141-91de-3a7cdbdc27c5))
      - {{embed ((65f1b8ff-f61b-49ff-86db-9177550b042b))}}
    - Personality Test collapsed:: true
    - Political Compass
    - Rejected
      - A/B Testing
        
        Can be biast towards the real thing
    - MaxDiff Analysis (Maximum Difference Scaling) id:: 65ea16cc-9983-47e6-acf8-0f71129e18c0
  - Qualitative background-color:: green collapsed:: true
    - DOING [[Focus Groups]]
    - LATER Delphi Method
    - Rejected collapsed:: true
      - Interviews
  - Unknown / Ideas background-color:: green collapsed:: true
    - Random background-color:: green collapsed:: true
      - https://conjointly.com/how-it-works/#step-4
      - Consumer panel
        
        TODO What is?
      - Product Concept Test
        
        link
      - Feature Preference Survey
        
        Conjoint Analysis
        
        ((65ea16cc-9983-47e6-acf8-0f71129e18c0))
    - TODO Monadic Testing collapsed:: true
      - Sequential Monadic Testing
      - Protomonadic Testing
    - TODO [[conjoint analysis]]
    - Testing [[MOM]] questions
    - Rejected collapsed:: true
      - Jobs-to-be-Done (JTBD)
      - Voice of the Customer (VOC)
      - Ethnographic Research
- ◾ LLMs Technicalities id:: 65ecd094-15df-41d3-b997-633ce3904ff2 collapsed:: true
  - ◾ Temperature id:: 6605ce4b-b28b-4001-a679-47c6565dccac
    - From paper collapsed:: true
      - The responses can be stochastic, depending on the “temperature” given to the model as a parameter. And, of course, different models can lead to different responses. Unlike the one homo economicus that is rational, there are many homo silci.
        
        From ((65ecc4f4-b5af-4f3b-b286-3b8787ec422e)) id:: 65ecd0b0-d7b9-4a77-b4e9-6dd8fb247aae
  - ◾ Models
- ◾ Personas id:: 65eccdd0-c8f6-4f02-9e1f-1f83b760a048 collapsed:: true
  - ◾ Persona Encoding id:: 66000105-916b-4588-b401-72b81acdd3e3
    - [[Big Five Personality Factors]] collapsed:: true
    - [[Sixteen Personality Factors (16PF)]]
  - ◾ Internal Consistency id:: 6601d330-7936-4411-a312-35e608c0beb1 collapsed:: true
    - [[Internal Consistency]]
  - {{query [[personas]]}} collapsed:: true
  - Random collapsed:: true
    - Alias
      - Characteristic Endowment
    - there is not a single LLM but rather a model capable of being conditioned to take on different personas that respond realistically
    - With a particular LLM, there is a single model, so it would seem that N = 1 no matter what. However, it has no fixed persona—it can be induced to play different agents via prompts.
      - From ((65ecd0b0-d7b9-4a77-b4e9-6dd8fb247aae)) id:: 65eccfc2-798f-472f-9a40-44a9bb08be59
- ◾ MR / Survey Prompt Engineering id:: 65ff2f20-285f-4826-97e1-12a518530483
- ◾ Disscussion id:: 65eccb27-ed38-419c-b16c-b896c8cde3d1 collapsed:: true
  - Weaknesses of LLMs collapsed:: true
    - “Stated versus revealed” preference critique, collapsed:: true
      - more
        
        ((65ecc4f4-b5af-4f3b-b286-3b8787ec422e))
      - ((65eccb87-3511-43d9-8ccc-4a5af9217bba))
    - Just a simulation collapsed:: true
      - more
        
        🔸 Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus? (Horton, 2023)
    - Other
      - ((65eccb0a-1363-4672-8032-5e8612220531))
        
        more
        
        ((65ecd0b0-d7b9-4a77-b4e9-6dd8fb247aae))
  - Framing
    - {{query [[framing]]}} collapsed:: true
    - Sometimes has a very minor to no effect
      - ((65ecd24e-8bf5-4cd0-bb1f-3a8807243350))
  - Stenghts
    - Clean within-subject experiment id:: 65ecd8ff-014b-4cf1-91a3-91af84fd84d9 collapsed:: true
      - collapsed:: true
        
        Compared to the original experiment, one benefit to the AI setting is that I can do a clean within-subject experiment because the AI does not “remember” having seen the previous prompt. In contrast, in a real experiment, a subject presented with the same scenario multiple times might get wise to the nature of the manipulation and alter results to make themselves more consistent.
        
        From ((65ecc4f4-b5af-4f3b-b286-3b8787ec422e))
- ◾ Data id:: 66000105-1fd1-456f-8ad4-12164a7e6b2c collapsed:: true
  - ◾ Benchmarking id:: 65ff2655-e2ac-4286-8215-d431240c1e7f
  - Way: Real focus groups collapsed:: true
    - Find focus group facilitator?
    - Find company willing to partner up – who has data on particular customers
  - Way: Big data collapsed:: true
    - Ideal: >1000 IRL people with characteristics and answers (benchmark model against those)
    - Characteristics --> Answers of IRL people
    - Comments from e.g., product hunt
    - 🔸 Comments on Youtube videos and Youtube video transcript id:: 65b6a0bf-1206-4ef6-8b01-c4326d0c2249
      - We can choose a video category
  - Way: Existing detailed market research data #new #discuss collapsed:: true
    - [[conjoint analysis]]
      - Ideas - Perhaps reach out to
        
        www.qualtrics.com
        
        https://sawtoothsoftware.com/
    - Public datasets
      - Sources
        
        TODO UCI Machine Learning Repository
        
        TODO Kaggle
        
        TODO Google Dataset Search
        
        Query for ' [[conjoint analysis]] '
        
        TODO go through
        
        tabs collapsed:: true
        
        https://www.kaggle.com/datasets/sachinsin8h/pizza-attributes-dataset-for-conjoint-analysis https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/6BSJYP https://scielo.figshare.com/articles/dataset/Using_conjoint_analysis_to_understand_customer_preferences_in_customized_low-income_housebuilding_projects/11350256 https://www.gov.uk/government/statistics/valuing-official-statistics-with-conjoint-analysis-april-2021 https://dataverse.harvard.edu/dataset.xhtml?persistentId=hdl:1902.1/22603 https://www.kaggle.com/datasets/dhruviskalpen/conjoint/code https://figshare.com/articles/dataset/CHOICE_OF_SALES_CHANNELS_IN_ELECTRONIC_COMMERCE/14286708/1 https://data.mendeley.com/datasets/8665r7htp4/2 https://data.mendeley.com/datasets/8wj5fsn3kt/1 https://zenodo.org/records/3560886 https://b2find.dkrz.de/dataset/440e7f06-aec2-5e58-ac60-d2f8a0be4f1c https://b2find.dkrz.de/dataset/21e968c3-a5d2-52d5-a459-ec0a034b78ed https://zenodo.org/records/8227363 http://datadiscoverystudio.org/geoportal/rest/metadata/item/39d4d1a5e06941a89aa403d6e92cd53f/html https://dataverse.harvard.edu/dataset.xhtml;jsessionid=22802515462841db26797b16b24f?persistentId=doi%3A10.7910%2FDVN%2F4WDVDB&version=&q=&fileTypeGroupFacet=&fileAccess= https://researchdata.edu.au/online-survey-2017-tourism-research/1440092 https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/03B2HW https://data.mendeley.com/datasets/96t8bknjc8/1 https://data.mendeley.com/datasets/zv6mf5w5m4/2 https://figshare.com/articles/dataset/Method_to_capture_and_prioritize_future_users_requirements_of_low-income_housing/20027393
        
        query: "Replication Data for"
      - List
        
        Pizza Attributes Dataset for Conjoint Analysis #discuss #kaggle id:: 65b69de8-338e-4a55-9c22-f497d24af9fb
        
        lnk
  - Way: Manual Assesment collapsed:: true
    - Manual assessment of output based on our intuition
    - Or from professional marketing professionals
  - ideas collapsed:: true
    - https://www.failory.com/failures
    - Any org. already know characteristics of their users/consumers
    - Alt.: Scrape SoMe and deduct characteristics based on user history. Answer (target-value)
- Assets/Resources
- Ideas
  - Replicating an extending ((66157a3e-4f6c-4490-bb7f-a4a51c3959b0))
    - Add Big5 persona encoding
  - Replicating and extending ((65af92bb-da88-45b8-90af-ca5770651146))
Writing collapsed:: true
- [[Methodology]]
  - TODO Define concepts collapsed:: true
    - {{embed ((662cc29e-a128-4062-b683-e981e2b24d3a))}}
- [[Literature Review]]
  - Frontiers of our Research id:: 6605ce4b-c3f7-49be-97a2-8bf1b720b1a6
    - LLMs Selection
    - Persona Encoding
    - Market Research Prompting
    - Evaluation
  - Random collapsed:: true
    - Go through lit and see how they encode the personas
  - Make a lit review collapsed:: true
    - Find and list existing papers
    - Make a draft'
      - List of papers
      - And bullet points.
    - See what can you replicate, simple, and then later we will build upon them.
    - We did this, we replicated these results, and tried to do this variation, cause we are marketing/business/prototype people
  - Observed patterns collapsed:: true
    - Internal consistency is common theme
      - refs
        
        {{embed ((660adaa7-49e2-44ab-b254-2dcecf94446a))}}
  - Relevant Assets
    - Thesis: Academic Papers
      - {{embed [[Thesis: Academic Papers]]}}
  - ❓ Including business applications collapsed:: true
    - {{embed ((6605ce4a-3d3c-4496-9c50-003ccfa7e4c1))}}
  - Learnings collapsed:: true
    - Weaknesses / Strenghts of Topic collapsed:: true
      - ((65ecc4f4-b5af-4f3b-b286-3b8787ec422e)) id:: 65ecce84-9d1d-4e95-9f58-c73eed441ce9
    - Input for Encodings collapsed:: true
      - We should avoid a textbook framing of questions.
        
        From ((65ecce84-9d1d-4e95-9f58-c73eed441ce9))
        
        2.3 The “performativity” problem
  - About background-color:: blue
    - [doing-a-literature-review.pdf](G:/My Drive/CBS/Thesis/doing-a-literature-review.pdf) #[[literature review]] id:: 662cca5d-9eff-4f0c-bb8d-bd9f649978db
      - How-to #TLDR background-color:: yellow
        
        🔹 Four tasks or sets of questions id:: 662ce1b3-6b6c-4bfd-9bde-606de410e3e2
        
        STEP 1 - What were the work trying to do? background-color:: green id:: 662ce1b7-851b-445d-bbfc-1fa434eefa19 collapsed:: true
        
        For each reviewed item
        
        What it was trying to do?
        
        was the dependent variable for the study?
        
        how was it conceptualized and operationalized?
        
        STEP 2 - Main argument of each work background-color:: green id:: 662ce3ed-89c5-4e9c-8e2c-b780c2b48b67 collapsed:: true
        
        For each reviewed item
        
        Was there a thesis?
        
        What is the argument?
        
        How strong is it?
        
        What qualifications or reservations does the author report?
        
        STEP 3 - Summarizing existing studies: 1) Consensus 2) Disagreement 3) Gaps background-color:: green
        
        1) Consensus
        
        2) Disagreement id:: 662ce4f9-d19d-4785-8e36-beb758b22aee collapsed:: true
        
        usually give rise to the alternative “camps” or “schools of thought” mentioned above. collapsed:: true
        
        {{embed ((662ce053-7562-4a92-bd53-a084a5b6d8c9))}}
        
        3) Gaps
        
        Relation to Step 4 collapsed:: true
        
        Expanding where the existing wisdom is less than conclusive.
        
        Potential flaws in the reasoning or evidence
        related to an area of consensus.
        
        Expanding on an existing debate is another possibility
        
        Proposing to fill a gap
        
        Utility collapsed:: true
        
        {{embed ((662ccfcb-a5cb-4e74-8650-9cd00872c474))}}
        
        STEP 4 - Assessing the quality of the literature and the overall state of knowledge on a topic background-color:: green id:: 662ce3f1-1a1c-4c19-857a-3aa5e35de350
        
        Assumptions collapsed:: true
        
        If there are disagreements, can they be traced to different assumptions made by the conflicting studies?
        
        How problematic are they?
        
        Logic collapsed:: true
        
        If there are disagreements, can they be traced to different theoretical perspectives?
        
        How problematic is the logic?
        
        Do the studies explain the reasoning that supports their key conclusions, or are important arguments made purely by assertion?
        
        What are the most plausible counterarguments or alternative explanations to the main thesis in each study, and does each study address these adequately?
        
        Is the reasoning that is provided logically persuasive, or does it contain internal contradictions
        
        Evidence collapsed:: true
        
        If there are disagreements, can they be traced to the use of different bodies of evidence
        
        Do the studies provide evidence to back up their main claims?
        
        Is the evidence valid—i.e., is it factually accurate and on point?
        
        Has all the relevant evidence been considered, or have some obviously relevant cases or bodies of data been overlooked?
        
        [[Methodology]]
        
        🔸 Structure collapsed:: true
        
        {{r not}}
        
        ❌ paragraph 1 notes that book A says X; paragraph 2 notes that article B says Y; paragraph 3 notes that book C says Z
        
        {{g do}}
        
        id:: 662cdecd-54cf-4993-ba0d-1d22ddf49616
        
        ‘A, B, and C argue that policy X has been ineffective and propose policy Y instead.’
        
        Categorize id:: 662ce053-7562-4a92-bd53-a084a5b6d8c9
        
        grouping individual studies into larger “camps” or “schools of thought.”
        
        in terms of
        
        different theories they propose or defend,
        
        different methodological approaches they take
        
        different policies they favor
        
        academics vs. government officials
        
        psychologists vs. economists
        
        🔸 Perhaps some scholars have already sought to classify the research id:: 662ce05c-7d20-48fc-baec-d7ba3f8b7512
        
        #[[Be the adult in the room]]
        
        By frontiers that we have
        {{embed ((6605ce4b-c3f7-49be-97a2-8bf1b720b1a6))}}
        
        Then, you can mention all three together in a single sentence such as
        
        ((662cdecd-54cf-4993-ba0d-1d22ddf49616))
      - Utility collapsed:: true
        
        find flaws in existing research
        
        new ideas on research
        
        not waste time “reinventing the wheel.”
        
        🔸 “lessons learned” collapsed:: true
        
        To determine and assess the practical know-how available
        
        in regard to which measures are likely to be effective or not
        
        enable you to place your research in a larger context collapsed:: true
        
        “contribution to knowledge,” or "value added"
        
        {{embed ((662ccfcb-a5cb-4e74-8650-9cd00872c474))}}
      - Details collapsed:: true
        
        Contribution to Knowledge id:: 662ccfcb-a5cb-4e74-8650-9cd00872c474 collapsed:: true
        
        Knowledge, in this context, does not necessarily mean “Truth” with a capital T.
        
        When reviewing literature, therefore, it is common to refer to the “claims” or “arguments” advanced by a study or school of thought.
        
        Idenfity the claims
        
        Asses the strength/support of these claims
        
        “researchers have studied a, b, and c, which are related to the problem of X, but they have not studied d, which is also relevant to understanding [or solving] X.”
        
        Not only academic work collapsed:: true
        
        In addition to acadmic
        
        policy dimension
        
        government agencies
        
        non-governmental organizations
        
        think tanks
        
        freelance researchers
        
        think of your task as a “review of existing knowledge"
      - Trivia collapsed:: true
        
        For each research study collapsed:: true
        
        Succinctly summarize the study’s {{g "main claim"}}
        
        the central argument - a sentence or two
        
        A literature review summarizes and evaluates a body of writings about a specific topic
        
        Two key elements
        
        Summarize the findings or claims from prior research on a topic. logseq.order-list-type:: number
        
        Reach a conclusion about how accurate and complete that knowledge is logseq.order-list-type:: number
        
        Focus on the body of work viewed as a while id:: 662cd15b-1980-479c-9508-febb964e4ed8
        
        your considered judgments about
        
        what’s right
        
        what’s wrong
        
        what’s inconclusive
        
        what’s missing
  - Plan background-color:: blue collapsed:: true
    - TODO Look into ((662ce05c-7d20-48fc-baec-d7ba3f8b7512))
  - CONTENT background-color:: purple
    - Step 1 and Step 2
      - Persona Encoding
        
        Focusing on
        
        Not focusing on
      - Market Research Prompting
        
        Focusing on
        
        Not focusing on
      - Evaluation
        
        Focusing on
        
        Not focusing on
      - Evaluation of the approach
Assets/Resources background-color:: blue id:: 65ecc30b-520a-4735-9af1-f85986f164b8 collapsed:: true
- Thesis: Academic Papers background-color:: yellow id:: 65eace94-0998-4aac-8934-5f317c07b37c
  - {{embed [[Thesis: Academic Papers]]}}
- Companies background-color:: yellow id:: 6605ce4a-3d3c-4496-9c50-003ccfa7e4c1
  - [[HN]]: Launch HN: Roundtable (YC S23) – Using AI to Simulate Surveys id:: 6605ce4b-63db-4bb4-a30c-4664bf8df28c collapsed:: true
    - src
    - Founder: Mayank Agrawal
  - 🔸 https://www.syntheticusers.com/
    - Use Cases
  - https://osum.com/?s=features id:: 65ff1597-4915-4727-b66f-3e9957a087c9
- Data background-color:: yellow
  - Data from consumer preference to utilise a mobile health app: a stated preference experiment
    - lnk
  - Electronic product adoption: consumer survey questionnaire, data, and interpreted results
    - LNK
- Other background-color:: yellow
  - VIDEO: Keynote: John Horton- Large Language Models as Economic Agents: What Can We Learn from Homo Silicus? collapsed:: true
    - TODO Watch video lnk
- Likely Irrelevant collapsed:: true
  - [[SimPy]] - discrete event simulation for Python
  - [[.embed]]{{embed ((65ecdee6-8ddf-4f95-8ee4-3a7d0e370e9a))}}
From RM collapsed:: true
- ((65e9bdba-b0ab-4133-9b6b-9fbaba222f98))
About Bachelor Thesis background-color:: purple
- Recommended literature collapsed:: true
  - Booth, Wayne C., Gregory G. Colomb, Joseph M. Williams, Joseph Bizup, and William T. FitzGerald. 2016. The craft of research. Chicago : The University of Chicago Press, 2016.
- Supervision collapsed:: true
  - 7 hours for 1 student (40 pages)
  - 12 hours for 2 students (60 pages)
- Formal collapsed:: true
  - - Formulate, delimit and analyze a problem within the framework of the DM program
  - - Select and adapt theories relevant to the problem and the analysis
  - - Substantiate the chosen method and research design
- 2 Studs: 60 pages, 1 stud: 40 pages
[[CBS Students Thesis]] background-color:: purple collapsed:: true
- {{embed [[CBS Students Thesis]]}}
Less Relevan background-color:: blue
- Consultations background-color:: purple collapsed:: true
  - [[Daniel]]
    - [[3_Daniel]] collapsed:: true
      - {{embed [[3_Daniel]]}}
    - Daniels meeting id:: 662cc29e-a128-4062-b683-e981e2b24d3a
      - Concept definitions
        
        Personas
        
        Market Research
    - From meeting
      - Make a shared documend
        
        Make notes from each meeting in it
        
        Notes from literature review
      - RAG Reasearch (retrieval) Augmented generation ...
[[Thesis: Irrelevant]] background-color:: blue collapsed:: true
- From older notes collapsed:: true
  - {{embed ((659da6d5-e9f8-4c13-80a9-d714c65cffb9))}}
  - {{embed ((65a6ca4a-fa9b-4dac-830e-7e5b881d8931))}}
- {{embed [[Thesis: Irrelevant]]}}
Syntax Description Test Text

Header Title Here's this

Paragraph Text

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Thesis.md

Thesis.md

Syntax	Description	Test Text
Header	Title	Here's this
Paragraph	Text

Files

Thesis.md

Latest commit

History

Thesis.md

File metadata and controls