ScanMyLove Analysis Methodology

Technical document for mental health professionals

Version: 5.0 Date: February 2026 Publisher: Psychologie et Sérénité Contact: contact@psychologieetserenite.com


Table of contents

  1. Introduction and epistemological framework
  2. Raw data analyzed (Inputs)
  3. The 14 clinical models used (Processing)
  4. Synthesis and insight generation (Outputs)
  5. Methodological limitations
  6. Scientific references

1. Introduction and epistemological framework

ScanMyLove is a computational analysis platform for written conversations (WhatsApp, Messenger, Instagram, Telegram, SMS, emails, screenshots) that applies 14 theoretical frameworks drawn from clinical psychology, social psychology and communication sciences to produce a relational map from textual exchanges.

The tool positions itself explicitly as a reflection-aid instrument and not as a diagnostic tool. It is a substitute neither for a clinical consultation nor for a standardized psychometric assessment. Its goal is to provide quantitative and qualitative benchmarks that help the user (and, where applicable, their therapist) better understand the relational dynamics at work in their exchanges.

Analysis pipeline architecture

The analysis process breaks down into three phases:

  1. Extraction and metrology (client side): parsing of conversations, computation of 25+ descriptive statistical metrics, detection of emotional-breakdown events.
  2. Multi-model analysis (server side): parallel processing by language models applying the 14 clinical grids to the anonymized corpus.
  3. Synthesis and reporting: aggregation of scores, production of composite dimensions, generation of the report with recommendations.

2. Raw data analyzed (Inputs)

2.1 Accepted data sources

Source Format Extraction method
WhatsApp Text export (.txt, .zip) FR/EN temporal regex (iOS/Android)
Messenger Facebook JSON export Structured JSON parsing
Instagram JSON export Structured JSON parsing
Telegram HTML/JSON export Multi-format parsing
SMS Text export Heuristic detection
Email Plain text Generic parsing
Screenshots PNG, JPG, WEBP, HEIC OCR via vision model (max 30 images)

Conversations are pre-processed for cross-file deduplication and encoding detection (UTF-8, UTF-16, Latin-1).

2.2 Descriptive metrics per participant

Volume and length

Metric Definition Unit
Number of messages Total messages sent Integer
Number of words Total words emitted Integer
Average length per message Words/messages ratio Words/message
Longest message Message with the highest word count Words

Linguistic and emotional richness

Metric Definition Computation method
Emoji density Total number of emojis used Unicode counting
Top 20 emojis Most frequent emojis per participant Descending frequency
Number of questions Messages containing "?" Punctuation detection
Number of exclamations Messages containing "!" Punctuation detection
Multimedia content Messages containing images, audio, video Tag detection
Top 30 words Most frequent words (excluding 150+ stopwords) Frequency after filtering
Top expressions Recurring bigrams and trigrams N-gram analysis

Temporality and responsiveness

Metric Definition Processing
Average response time Mean delay between received message and reply Arithmetic mean, filtering >24h
Median response time Median value of delays Median, more robust to outliers
Hourly distribution Message distribution across 24 hours Histogram 0-23h
Weekly distribution Distribution across 7 days Sunday to Saturday
Activity heatmap 7 days x 24 hours matrix Message density per time slot
Longest active streak Maximum number of consecutive days with exchange Streak counting

Dynamics indicators

Metric Definition
Exchange initiator Identification of who sends the first message of each daily conversation
Global first message Author of the very first message of the corpus
Global last message Author of the most recent message
Total conversation duration Number of days covered by the corpus
Messages per day (average) Average daily volume

2.3 Detection of emotional-breakdown events

The breakpoint engine detects 5 types of events through a longitudinal analysis of the corpus.

Affective ratio (7-day sliding window)

The computation relies on two dictionaries of linguistic markers:

Formula:

affective_ratio = positive_markers / (positive_markers + negative_markers + 1)

The ratio ranges between 0 (entirely negative) and 1 (entirely positive). Smoothing over 7 days reduces the noise from daily fluctuations.

Types of events detected

Event Detection criterion Severity
Affective drop Affective ratio decline >30% (medium) or >60% (high) over 14 days Medium / High
Ghosting Silence of 3-6 days (medium) or 7+ days (high) after 5+ active days Medium / High
Communication withdrawal Weekly volume decline >60% (medium) or >80% (high) Medium / High
Acute conflict Spike of negative markers >3x the daily average Medium / High
Composite breakup Co-occurrence of an affective drop and ghosting within a 14-day window Always High

Events are deduplicated within a 3-day window to avoid redundancy.


3. The 14 clinical models used (Processing)

Processing architecture

The clinical analysis is performed by a parallel processing pipeline organized into groups. Each group independently queries a language model with a specific prompt encoding a structured assessment grid and a JSON output schema.

Two levels of analysis are offered:

Level Groups Models covered Average duration
Standard (€9.90) 5 parallel groups 7 models ~22 seconds
Premium (€19.90) 9 parallel groups 14 models + action plan ~33 seconds

The conversational corpus is first anonymized (names replaced by "Person A" / "Person B"; phone numbers, emails, addresses and banking details redacted) and truncated according to a 160,000-character budget strategically allocated:

This allocation ensures the model has access to the initial context, the intermediate evolution and the current state of the relationship.


Model 1: Sternberg's Triangular Theory of Love

Theoretical reference: Sternberg, R. J. (1986). A Triangular Theory of Love. Psychological Review, 93(2), 119-135.

What it measures: Three fundamental components of love — intimacy (emotional closeness, trust, sharing), passion (desire, attraction, enthusiasm) and commitment (conscious decision to maintain the relationship).

Scoring method: Each component is assessed on a 0-to-100 scale from the analysis of conversational markers:

Clinical interpretation: The combination of the three scores produces a typology of 7 forms of love (consummate love, companionate love, infatuation, empty love, romantic love, fatuous love, liking/friendship).


Model 2: The 5 Love Languages (Chapman)

Theoretical reference: Chapman, G. (1992). The Five Love Languages: How to Express Heartfelt Commitment to Your Mate. Northfield Publishing.

What it measures: Each participant's preferred mode of expressing and receiving affection.

Scoring method: Five dimensions assessed from 0 to 100 for each participant:

Clinical interpretation: The analysis identifies each participant's dominant language and measures the degree of concordance or mismatch (gap analysis). A significant gap between the two participants' primary languages is an indicator of potential relational frustration.


Model 3: The Four Horsemen of the Apocalypse (Gottman)

Theoretical reference: Gottman, J. M. (1994). What Predicts Divorce? The Relationship Between Marital Processes and Marital Outcomes. Lawrence Erlbaum Associates. Gottman, J. M., & Silver, N. (1999). The Seven Principles for Making Marriage Work. Three Rivers Press.

What it measures: The four communication behaviors most predictive of couple dissolution according to the work of the Gottman Institute.

Scoring method: Four indicators assessed from 0 to 100:

Positive/negative ratio: Computation of the ratio of positive vs. negative interactions.

Ratio Interpretation
>= 5:1 Healthy relationship (Gottman threshold)
3:1 to 5:1 Watch zone
< 3:1 Alert zone (one horseman > 50)
< 1:1 Critical zone (one horseman > 70)

Clinical interpretation: Gottman's research indicates that the simultaneous presence of these four behaviors predicts separation with roughly 93% accuracy over 6 years. Contempt is considered the most toxic predictor.


Model 4: Power dynamics (Pursuer / Distancer)

Theoretical reference: Johnson, S. M. (2008). Hold Me Tight: Seven Conversations for a Lifetime of Love. Little, Brown and Company. Fogarty, T. F. (1976). Marital crisis. In P. J. Guerin (Ed.), Family Therapy: Theory and Practice. Gardner Press.

What it measures: The distribution of emotional and communicational power in the couple, measured through the pursuer/distancer dynamic.

Scoring method:

Imbalance levels:

Gap Classification
0-15 Balance
16-30 Slight imbalance
31-50 Imbalance
> 50 Strong imbalance

Clinical interpretation: A pursuer/distancer imbalance is self-reinforcing: the more the pursuer insists, the more the distancer withdraws, which increases the pursuer's anxiety (Johnson, 2008). Identifying this dynamic is a key step in Emotionally Focused Couple Therapy (EFT).


Model 5: Attachment styles (Bowlby, Ainsworth, Bartholomew)

Theoretical references: Bowlby, J. (1969). Attachment and Loss, Vol. 1: Attachment. Basic Books. Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of Attachment. Lawrence Erlbaum. Bartholomew, K., & Horowitz, L. M. (1991). Attachment Styles Among Young Adults: A Test of a Four-Category Model. Journal of Personality and Social Psychology, 61(2), 226-244.

What it measures: Each participant's predominant attachment style, inferred from their conversational behaviors.

Scoring method: Four dimensions assessed from 0 to 100 for each participant:

Clinical interpretation: Attachment style is the best predictor of recurring relational patterns (Hazan & Shaver, 1987). The analysis identifies the dominant style and measures the compatibility of the two participants' styles.


Model 6: Cognitive distortions (Beck)

Theoretical reference: Beck, A. T. (1976). Cognitive Therapy and the Emotional Disorders. International Universities Press. Burns, D. D. (1980). Feeling Good: The New Mood Therapy. William Morrow.

What it measures: The systematic information-processing errors present in each participant's discourse.

Scoring method: 10+ types of distortions assessed from 0 to 100, with identification of illustrative excerpts:

Distortion Definition Typical conversational example
Mind reading Presuming to know the other's intentions "You say that but I know you think the opposite"
Personalization Attributing to oneself the cause of external events "If you're in a bad mood it's because of me"
Overgeneralization Drawing a general rule from an isolated case "You NEVER make any effort"
Catastrophizing Imagining the worst-case scenario "If you don't reply it means it's over"
Emotional reasoning Confusing emotion with reality "I feel it, so it's true"
Mental filter Retaining only the negative Ignoring compliments, focusing on one remark
Dichotomous thinking All or nothing "Either you love me completely or you don't love me"
Labeling Reducing the other to a single trait "You're a selfish person"
Minimization Devaluing the positive "OK you did that, but it's normal"
Imperatives ("should") Imposing rigid norms "You should know what I need"

Clinical interpretation: The frequency and diversity of cognitive distortions are correlated with relational distress (Epstein & Baucom, 2002). Identifying them is a major therapeutic lever in couple CBT. For each distortion detected, 1 to 2 real excerpts from the conversation are provided.


Model 7: Young's Early Maladaptive Schemas (Premium)

Theoretical reference: Young, J. E., Klosko, J. S., & Weishaar, M. E. (2003). Schema Therapy: A Practitioner's Guide. Guilford Press.

What it measures: The Early Maladaptive Schemas activated in the relational dynamic, assessed for each participant.

Scoring method: 18 schemas assessed from 0 to 100, grouped into 5 domains:

Domain 1 — Disconnection and rejection

Domain 2 — Impaired autonomy and performance

Domain 3 — Impaired limits

Domain 4 — Other-directedness

Domain 5 — Overvigilance and inhibition

Clinical interpretation: Early maladaptive schemas are stable, self-perpetuating patterns that bias the interpretation of relational events (Young et al., 2003). Detecting them in textual exchanges helps identify the core schema themes that organize the couple's dynamic.


Model 8: Nonviolent Communication (Premium)

Theoretical reference: Rosenberg, M. B. (2003). Nonviolent Communication: A Language of Life. PuddleDancer Press.

What it measures: Each participant's ability to communicate according to NVC's OFNR process (Observation, Feeling, Need, Request).

Scoring method: Global score from 0 to 100 per participant, broken down into:

Clinical interpretation: Message excerpts are provided with an NVC / non-NVC analysis and proposed reformulations. NVC is widely used in couple mediation and relational therapy.


Model 9: DISC profile (Premium)

Theoretical reference: Marston, W. M. (1928). Emotions of Normal People. Kegan Paul. Adapted and modernized by J. Geier (1958) and Wiley's DiSC model.

What it measures: Each participant's behavioral communication style along four axes.

Scoring method: Four dimensions assessed from 0 to 100 per participant:

Clinical interpretation: DISC incompatibilities (e.g., high D + high S in the other) are sources of communicational friction. Identifying the profile helps adapt dialogue strategies.


Model 10: Karpman's Drama Triangle (Premium)

Theoretical reference: Karpman, S. B. (1968). Fairy Tales and Script Drama Analysis. Transactional Analysis Bulletin, 7(26), 39-43.

What it measures: The three dysfunctional roles adopted in relational psychological games.

Scoring method: Three roles assessed from 0 to 100 per participant:

Clinical interpretation: The analysis also detects role shifts over time. The circularity of Karpman's games is a key concept in transactional analysis: the persecutor can become a victim, the victim can turn into a persecutor, and so on.


Model 11: The Johari Window (Premium)

Theoretical reference: Luft, J., & Ingham, H. (1955). The Johari Window: A Graphic Model of Interpersonal Awareness. Proceedings of the Western Training Laboratory in Group Development, UCLA.

What it measures: Each participant's degree of openness and transparency in the relationship.

Scoring method: Four zones assessed from 0 to 100 per participant:

Clinical interpretation: A large open area is associated with better relational satisfaction. The analysis identifies the degree of reciprocity in self-disclosure.


Model 12: Schutz's FIRO (Premium)

Theoretical reference: Schutz, W. (1958). FIRO: A Three-Dimensional Theory of Interpersonal Behavior. Holt, Rinehart & Winston.

What it measures: The three fundamental interpersonal needs, in their expressed and wanted dimensions.

Scoring method: 6 scores from 0 to 100 per participant (3 needs x 2 dimensions):

Need Expressed Wanted
Inclusion Tendency to include the other Need to be included
Control Tendency to lead Need to be guided
Affection Tendency to express affection Need to receive affection

Clinical interpretation: The gap between expressed and wanted scores reveals unverbalized interpersonal frustrations. A large gap on the affection axis (strong wanted need, weak expression from the other) is an indicator of emotional deprivation.


Model 13: Polyvagal Theory (Premium)

Theoretical reference: Porges, S. W. (2011). The Polyvagal Theory: Neurophysiological Foundations of Emotions, Attachment, Communication, and Self-Regulation. W. W. Norton.

What it measures: The state of the autonomic nervous system inferred from the communication style, according to the three branches of the vagus nerve.

Scoring method: Three states assessed from 0 to 100 per participant:

Clinical interpretation: Polyvagal theory offers a neurobiological framework for understanding withdrawal or aggression reactions in the couple. A participant in a prolonged dorsal state requires a different therapeutic approach from one in sympathetic activation.


Model 14: Violence detection (Premium)

This model breaks down into two complementary sub-models, activated conditionally.

14a. Walker's Cycle of Violence

Theoretical reference: Walker, L. E. (1979). The Battered Woman. Harper & Row. Walker, L. E. (2009). The Battered Woman Syndrome (3rd ed.). Springer Publishing.

Trigger: Activated if the global toxicity score > 60, or if more than 3 affective drops or ghosting episodes are detected.

What it measures: The presence and phases of the cycle of domestic violence.

Scoring method: Identification of the 4 phases with dating:

  1. Tension phase: stress build-up, growing irritability
  2. Explosion phase: acting out (verbal in the textual context)
  3. Reconciliation phase: apologies, promises of change
  4. Honeymoon phase: return to affection, idealization

14b. Power and Control Wheel (Duluth)

Theoretical reference: Pence, E., & Paymar, M. (1993). Education Groups for Men Who Batter: The Duluth Model. Springer Publishing.

Trigger: Activated if the toxicity score > 50.

What it measures: The 8 power-and-control mechanisms identified by the Duluth model.

Scoring method: Eight indicators assessed from 0 to 100:

Clinical interpretation: These two models are activated conditionally to avoid false positives in healthy relationships. When activated, a safety banner with emergency numbers (in France: 3114 — suicide prevention, 3919 — domestic violence, 114 — emergencies by SMS) is displayed prominently.


4. Synthesis and insight generation (Outputs)

4.1 Composite dimensions

The individual scores of the 14 models are aggregated into composite dimensions that offer a synthetic view of the relationship.

Standard level (5 dimensions):

Dimension Contributing models
Affection Sternberg (intimacy, passion), Chapman (languages), FIRO (affection)
Communication Gottman (ratio), NVC, DISC, power dynamics
Conflict Gottman (4 horsemen), Karpman, cognitive distortions
Commitment Sternberg (commitment), shared plans, prospective discourse
Balance Power dynamics, attachment, reciprocity of exchanges

Premium level (additional dimensions):

Dimension Contributing models
Trust Attachment, Young (mistrust/abuse), Johari (open area)
Intimacy Sternberg, Johari, FIRO (inclusion/affection), self-disclosure
Respect Gottman (absence of contempt), Duluth, power balance
Humor Detection of lightness, wordplay, playful emojis
Projection Future orientation, planning, explicit commitment

4.2 Global score and levels

The global score is computed as the arithmetic mean of all dimensions.

Global score Interpretation
>= 70 / 100 Positive relationship
40 - 69 / 100 Relationship under tension
< 40 / 100 Relationship in difficulty

4.3 Relational prognosis

The prognosis is generated from all indicators and includes:

4.4 Therapeutic action plan (Premium)

The Premium report includes an action plan structured in 3 phases, each containing 2 to 3 concrete exercises with title, description and recommended frequency.


5. Methodological limitations

5.1 What the analysis CANNOT do

  1. Make a clinical diagnosis. ScanMyLove is not a diagnostic tool in the sense of the DSM-5 or ICD-11. It cannot diagnose a personality disorder, depression, an anxiety disorder or any other pathology.

  2. Substitute for a mental health professional. Computational text analysis does not replace clinical observation, face-to-face interviews, standardized psychometric tests and the clinician's expertise.

  3. Analyze non-verbal communication. Paralanguage (tone, pace, volume), facial expressions, body language and the physical context of the exchanges inherently escape textual analysis.

  4. Assess the sincerity of the discourse. The analysis concerns what is written, not what is thought. A participant may use affectionate language out of manipulation or social convention.

  5. Capture off-text exchanges. Face-to-face conversations, phone calls and non-digital interactions are not taken into account. The analysis concerns a potentially unrepresentative sample of the couple's total communication.

5.2 Identified biases

Bias Description Mitigation
Sampling bias The analyzed conversation may not be representative of the relationship Explicit mention in the report
Social desirability bias Written exchanges may be more "filtered" than oral communication Taken into account in interpretation
Cultural bias The affective-marker dictionaries are calibrated on metropolitan French Progressive extension to other variants
Recency bias Overrepresentation of recent messages (50% of the token budget) Deliberate choice, as the current state is clinically the most relevant
Length bias Short conversations produce less reliable analyses Automatic warning below a message threshold
Platform bias Writing style varies by platform (SMS vs. email) Platform detection and contextual adjustment
Model-interpretation bias Language models may hallucinate or over-interpret Structured prompts with strict JSON schemas and caution instructions

5.3 Ethical positioning


6. Scientific references

Attachment theory

Couple research (Gottman)

Triangular theory of love

Love languages

Schema therapy

Cognitive therapy and distortions

Nonviolent Communication

DISC model

Transactional analysis and Karpman's triangle

Johari Window

FIRO

Polyvagal theory

Domestic violence

Emotionally Focused Couple Therapy

Pursuer-distancer dynamics


Appendix: Summary of models and analysis levels

# Model Primary reference Level Scores
1 Triangular Theory of Love Sternberg (1986) Standard 3 x (0-100)
2 5 Love Languages Chapman (1992) Standard 5 x (0-100) / participant
3 Gottman's 4 Horsemen Gottman (1994) Standard 4 x (0-100) + ratio
4 Power dynamics Johnson (2008), Fogarty (1976) Standard 2 x (0-100) + balance
5 Attachment styles Bowlby (1969), Bartholomew (1991) Standard 4 x (0-100) / participant
6 Cognitive distortions Beck (1976), Burns (1980) Standard 10+ x (0-100) + excerpts
7 Young's schemas Young et al. (2003) Premium 18 x (0-100) / participant
8 Nonviolent Communication Rosenberg (2003) Premium 1 x (0-100) / participant
9 DISC profile Marston (1928) Premium 4 x (0-100) / participant
10 Karpman's triangle Karpman (1968) Premium 3 x (0-100) / participant
11 Johari Window Luft & Ingham (1955) Premium 4 x (0-100) / participant
12 Schutz's FIRO Schutz (1958) Premium 6 x (0-100) / participant
13 Polyvagal theory Porges (2011) Premium 3 x (0-100) / participant
14 Violence detection Walker (1979), Pence & Paymar (1993) Premium 4 phases + 8 x (0-100)

Document produced from analysis of the ScanMyLove v5.0 source code. This document is intended for mental health professionals evaluating the credibility and methodological rigor of the tool. For any question or collaboration, contact contact@psychologieetserenite.com.